AU2012375204A1 - Image analysis for determining characteristics of individuals and group of individuals - Google Patents

Abstract

Embodiments include methods for predicting one or more characteristics of an individual, such as a human or non-human animal, by applying computational methods to image(s) of the individual to generate one or more metrics indicative of the characteristics. Embodiments determine predictors of characteristics by creating a sample library of individuals, determining facial measurements for each individual, determining relationships between facial measurements and additional library data, and selecting predictors from these relationships. Embodiments include methods for predicting characteristics of individuals not in the library. Embodiments include methods for predicting characteristics of groups using predicted characteristics of individuals. Embodiments determine suitability of a pair of individuals (from the same or different species) for a particular purpose, task, or relationship based on characteristics of individuals. Other embodiments determine the compatibility of an individual with a group of other individuals. Embodiments include systems, devices, and computer-readable media comprising one or more of these methods.

Description

WO 2013/147932 PCT/US2012/064178 IMAGE ANALYSIS FOR DETERMINING CHARACTERISTICS OF INDIVIDUALS AND GROUPS OF INDIVIDUALS CROSS-REFERENCE TO RELATED APPLICATIONS 5 [0001] This application claims the priority of U.S. Provisional Application Serial No. 61/616,234, entitled "Image Analysis for Determining Characteristics of Animals," which was filed March 27, 2012. This application also claims the priority of U. S. Application Serial No. 13/467,869, entitled "Image Analysis for Determining Characteristics of Animals," and U. S. Application Serial No. 13/467,889, entitled "Image Analysis for 10 Determining Characteristics of Humans," both of which were filed May 9, 2012. TECHNICAL FIELD [0002] The disclosure herein relates to the objective determination of a characteristic of an individual, such as a human or non-human animal, by applying computational methods to one or more images of the individual to generate one or more metrics indicative of the 15 characteristic of interest. It also relates to the determining the degree to which particular groups of two or more individuals are suitable for a particular purpose, task, or relationship. BACKGROUND [0003] Animal domestication can be thought of as developing a mutually useful 20 relationship between animals and humans. Over the past 12,000 years, humans have learned to control their access to food and other necessities of life by changing the behaviors and natures of wild animals. All of today's domesticated animals - including dogs, cats, cattle, oxen, llamas, sheep, goats, camels, geese, horses, chickens, turkeys, and pigs - started out as wild animals but were changed over the centuries and millennia into 25 animals that are tamer, quieter, and generally more cognitively suited to a lifestyle of coexistence with humans. Today people benefit from domesticated animal in many ways including keeping cattle in pens for access to milk and meat and for pulling plows, training dogs to be guardians and companions, teaching horses to adapt to the plow or take a rider, and changing the lean, nasty wild boar into the fat, friendly pig.

WO 2013/147932 PCT/US2012/064178 2 [0004] When individuals are looking to breed animals, they look for certain traits in purebred stock that are valued for a particular purpose, or may intend to use some type of crossbreeding to produce a new type of stock with different, and, it is presumed, superior abilities in a given area of endeavor. For example, to breed chickens, a typical breeder 5 intends to receive eggs, meat, and new, young birds for further reproduction. Thus, the breeder has to study different breeds and types of chickens and analyze what can be expected from a certain set of characteristics before he or she starts breeding them. On the other hand, purebred breeding aims to establish and maintain stable traits that animals will pass to the next generation. By "breeding the best to the best," employing a certain 10 degree of inbreeding, considerable culling, and selection for "superior" qualities, one could develop a bloodline superior in certain respects to the original base stock. [0005] As first noted by Charles Darwin, domesticated animals are known to share a common set of physical characteristics, sometimes referred to as the domestication phenotype. C. Darwin, THE VARIATION OF ANIMALS AND PLANTS UNDER 15 DOMESTICATION (2"d ed.) (New York: D. Appleton & Co., 1883). They are often smaller, with floppier ears and curlier tails than their untamed ancestors. Their coats are sometimes spotted while their wild ancestors' coats are solid. One long-term study demonstrating this phenomenon has been ongoing since 1958 at the Institute of Cytology and Genetics in Novosibirsk, Russia. In this study, scientists have successfully 20 demonstrated that, through careful selective breeding for tamability, wild Siberian silver foxes acquire both the behavioral and appearance traits of domesticated dogs. See, e.g., L. Trut, Early Canid Domestication: The Fox Farm Experiment, 87 AMERICAN SCIENTIST 160-69 (Mar.-Apr. 1999). This highly conserved combination of psychological and morphological changes during the process of domestication is seen to varying degrees 25 across a remarkably wide range of species, from horses, dogs, pigs, and cows to some non-mammals like chickens and even a few fish. However, in no other species has this relationship between behavior and anatomical features been more widely noted than in the horse. [0006] Relationships also exist in humans between physiological feature sets (i.e., 30 phenotypes) and certain cognitive functions and/or personality traits. During progressive stages of human embryonic growth, development of the brain and face remains intimately WO 2013/147932 PCT/US2012/064178 3 connected through both genetic signaling and biomechanical/biochemical mechanisms. The face develops from populations of cells originating from the early neural crest, with cells from the neural tube gradually shifting to form the prominences of the face. Differentiation of these early cells is closely regulated through intricate genetic signaling 5 mechanisms, with the brain essentially serving as the platform on which the face grows. As these two structures continue to grow and develop during the later embryonic stages, their phenotypes remain closely linked as complex genetic hierarchies regulate patterns of cross talk between molecules, cells, and tissues. SUMMARY 10 [0007] Embodiments comprise a method for measuring an individual, such as a human or non-human animal, to determine one or more characteristics of the individual, comprising receiving one or more digital images representing the individual, storing the images in a computer memory, adding a plurality of reference points to the stored digital images, and computing one or more metrics relating to the characteristic of the individual 15 using the reference points. Other embodiments comprise a method for determining a characteristic of an individual based on a set of metrics related to the individual, comprising selecting one or more metrics from the set of metrics, calculating a combined metric using the selected metrics, and determining the characteristic of the individual based on the value of the combined metric. Other embodiments comprise computer 20 systems that implement one or more of the above methods. [0008] Other embodiments include a method for predicting a human characteristic comprising, for each of a plurality of individuals, storing one or more digital images representing the individual in a memory operably connected to a digital computer; annotating the one or more digital images with a plurality of reference points; associating 25 at least one other data value about the individual with the one or more digital images representing the individual; computing, with the digital computer, one or more metrics using the plurality of reference points. The method further comprises selecting a combination of the one or more metrics for predicting the human characteristic. In some embodiments, the selecting step further comprises determining one or more relationships 30 between the one or more metrics and the at least one other data value for the plurality of WO 2013/147932 PCT/US2012/064178 4 individuals and the combination is selected based on the one or more relationships. Other embodiments comprise systems and computer-readable media embodying these methods. [0009] Other embodiments include a method for determining a characteristic of an individual human based on one or more metrics related to the individual, comprising 5 storing one or more digital images representing the individual in a memory operably connected to a digital computer; annotating the one or more digital images with a plurality of reference points; computing, with the digital computer, the one or more metrics using the plurality of reference points; and predicting the characteristic of the individual based on the one or more metrics. In some embodiments, the predicting step 10 further comprises computing a combined metric based on a predetermined function of the one or more metrics and predicting the characteristic based on the combined metric. In some embodiments, predicting the one or more characteristics of the individual based on the one or more metrics comprises computing one or more indicators used for predicting the one or more characteristics. Depending on embodiment, the one or more indicators 15 may comprise one or more primary indicators, one or more secondary indicators, one or more tertiary indicators, and/or one or more suitability indicators. Other embodiments comprise systems and computer-readable media embodying these methods. [0010] Other embodiments include a method for predicting one or more characteristics of a group of two or more individuals comprising, for each of the individuals within the 20 group, storing one or more digital images representing the individual in a memory operably connected to a digital computer; annotating the one or more digital images with a plurality of reference points; computing, with the digital computer, one or more metrics using the plurality of reference points; and predicting one or more characteristics of the individual based on the one or more metrics. The method further comprises predicting 25 the one or more characteristics of the group of individuals based on the predicted one or more characteristics of the individual comprising the group. Some embodiments further comprise determining a strategy relating to the group of individuals based on the predicted one or more characteristics of the group. Some embodiments further comprise computing one or more combined metrics, each based on a predetermined function of the 30 one or more metrics, and predicting the one or more characteristics of an individual based on the one or more combined metrics. In some embodiments, the group comprises a pair WO 2013/147932 PCT/US2012/064178 5 of individuals, and predicting the one or more characteristics of the pair based on the one or more metrics comprises computing one or more indicators used for predicting the one or more characteristics. In some embodiments, the one or more indicators may comprise one or more individual indicators (e.g., principal, secondary, and/or tertiary indicators) 5 and one or more pairwise indicators related to a pair of individuals. Other embodiments comprise systems and computer-readable media embodying these methods. [0011] Other embodiments include a method for determining a characteristic of a subject (e.g., a human), comprising calculating two or more ratios based upon metrics related to a subject's head, wherein distances or angles between reference points on the 10 subject's head are used; predicting, using a computer and computations, a characteristic of the subject wherein the two or more ratios are used and wherein data about a group of subjects are referenced; and providing the predicted characteristic to an output device. Other embodiments comprise systems and computer-readable media embodying these methods. 15 [0012] Other embodiments include a method for choosing a combination of two individuals for a particular task, comprising computing one or more metrics related to one individual; computing one or more metrics related to each of a plurality of second individuals; computing a combination characteristic related to the combination of the first individual with each of the plurality of second individuals, based on at least a portion of 20 the one or more metrics related to the first individual and at least a portion of the one or more metrics related to each of the plurality of second individuals; and determining the combination of the first individual and one of the plurality of second individuals based on the computed combination characteristics. Depending on embodiment, the first individual may be a human or non-human animal and all of the plurality of second 25 individuals may be humans or members of the same non-human animal species. Other embodiments comprise systems and computer-readable media embodying these methods. DESCRIPTION OF THE DRAWINGS [0013] The detailed description will refer to the following drawings, wherein like numerals refer to like elements, and wherein: 30 Fig. 1A shows facial descriptor measurement Eye Height Proportion (EHP); WO 2013/147932 PCT/US2012/064178 6 Fig. 1B shows facial descriptor measurement Eye Depth Proportion (EDP); Fig. IC shows facial descriptor measurement Maximal Eye Point Proportion (MaxEPP); Fig. ID shows facial descriptor measurement Minimal Eye Point Proportion (MinEPP); Fig. 1E shows facial descriptor measurement Eye Height to Length Proportion (EHLP); 5 Fig. IF shows facial descriptor measurement Eye Extrema Angle (EEA); Fig. 1G shows facial descriptor measurement Upper Lateral Eye Roundness Proportion (ULERP); Fig. 1H shows facial descriptor measurement Lower Lateral Eye Roundness Proportion (LLERP); 10 Fig. 1J shows facial descriptor measurement Upper Rostral Eye Roundness Proportion (URERP); Fig. 1K shows facial descriptor measurement Lower Rostral Eye Roundness Proportion (LRERP); Fig. IL shows facial descriptor measurement Pupilary Area Ratio (PAR), which is the 15 area the area covered by the pupil when maximally dilated, characterized by radially striated bands around the area when not fully dilated; Fig. 1M shows facial descriptor measurement Eye Set Angle (ESA); Fig. 1N shows facial descriptor measurement Eye Width Proportion (EWP); Fig. IP shows facial descriptor measurement Iris Boundary Proportion (IBP); 20 Fig. 1Q shows facial descriptor measurement Caruncle Length Proportion (CLP); Fig. iR shows facial descriptor measurement Caruncle Angular Proportion (CAP); Fig. 1S shows facial descriptor measurement Upper Eye Angle (UEA); Fig. 1T shows facial descriptor measurement Lower Eye Angle (LEA); Fig. 1U shows facial descriptor measurement Front Eye Angle (FEA); 25 Fig. 1V shows facial descriptor measurement Eye Elevation Angle (EEA); Fig. 1W shows facial descriptor measurement Eye Depression Angle (EDA); Fig. IX shows facial descriptor measurement Eye Set Depth Proportion (ESDP); Fig. 2A shows facial descriptor measurement Maxillary Palate Angle (MPA); Fig. 2B shows facial descriptor measurement Maxillary Plate Prominence Proportion 30 (MPPP); Fig. 2C shows facial descriptor measurement Facial Length Proportion (FLP); WO 2013/147932 PCT/US2012/064178 7 Fig. 2D shows facial descriptor measurement Mid-Face Proportion (MFP); Fig. 2E shows facial descriptor measurement Mandible Width Proportion (MWP); Fig. 2F shows facial descriptor measurement Labio-Orbital Angle (LOA); Fig. 2G shows facial descriptor measurement Naso-Orbital Angle (NOA); 5 Fig. 2H shows facial descriptor measurement Naso-Chilial Angle (NCA);Fig. 3A shows facial descriptor measurement Upper Lip Protuberance Proportion (ULPP); Fig. 3B shows facial descriptor measurement Lower Lip Protuberance Proportion (LLPP); Fig. 3C shows facial descriptor measurement Upper Lip Thickness Proportion (ULTP); 10 Fig. 3D shows facial descriptor measurement Lower Lip Thickness Proportion (LLTP); Fig. 3E shows facial descriptor measurement Lower Lip Maxima Point Proportion (LLMPP); Fig. 3F shows facial descriptor measurement Upper Lip Maxima Point Proportion (ULMPP); 15 Fig. 3G shows facial descriptor measurement Lower Lip Angle (LLA); Fig. 3H shows facial descriptor measurement Upper Lip Angle (ULA); Fig. 3J shows facial descriptor measurement Lower Lip Roundness Proportion (LLRP); Fig. 3K shows facial descriptor measurement Upper Lip Roundness Proportion (ULRP); Fig. 3L shows facial descriptor measurement Lip Convergence Angle (LCA); 20 Fig. 3M shows facial descriptor measurement Mouth-to-Pupil Width Proportion (MtPWP); Fig. 3N shows facial descriptor measurement Mouth-to-Nostril Width Proportion (MtNWP); Fig. 3P shows facial descriptor measurement Mouth Thickness Proportion (MTP); 25 Fig. 3Q shows facial descriptor measurement Mouth Width Angle (MWA); Fig. 3R shows facial descriptor measurement Upper Lip Length Proportion (ULLP); Fig. 3S shows facial descriptor measurement Lower Lip Length Proportion (LLLP); Fig. 3T shows facial descriptor measurement Upper Labial Thickness Proportion (ULTP); Fig. 3U shows facial descriptor measurement Lower Labial Thickness Proportion 30 (LLTP); Fig. 4A shows facial descriptor measurement Jowl Fullness Proportion (JFP); WO 2013/147932 PCT/US2012/064178 8 Fig. 4B shows facial descriptor measurement Jowl Connective Proportion (JCP); Fig. 4C shows facial descriptor measurement Jowl Thickness Angle (JTA); Fig. 4D shows facial descriptor measurement Jowl Set Angle (JSA); Fig. 4E shows facial descriptor measurement Lambiomental Inflexion Proportion (LIP); 5 Fig. 4F shows facial descriptor measurement Chin Roundness Proportion (CRP); Fig. 4G shows facial descriptor measurement Chin Height Proportion (CHP); Fig. 4H shows facial descriptor measurement Jowl Depth Angle (JDA); Fig. 5D shows facial descriptor measurement Radix Protrusion Proportion (RPP); Fig. 5E shows facial descriptor measurement Radix Protrusion Angle (RPA); 10 Fig. 5F shows facial descriptor measurement Dorsum Rounding Proportion (DRP); Fig. 5G shows facial descriptor measurement Dorsum Maximal Protuberance Point Proportion (DMPPP); Fig. 5H shows facial descriptor measurement Nose Rounding Proportion (NRP); Fig. 5J shows facial descriptor measurement Nostril Tip Rounding Proportion (NTRP); 15 Fig. 5K shows facial descriptor measurement Nasal Tip Angle (NTA); Fig. 5L shows facial descriptor measurement Nostril Height Proportion (NHP); Fig. 5M shows facial descriptor measurement Nares Protrusion Proportion (NaPP); Fig. 5N shows facial descriptor measurement Nares Length Proportion (NaLP); Fig. 5P shows facial descriptor measurement Nares Inflexion Proportion (NIP); 20 Fig. 5Q shows facial descriptor measurement Full Nose Angle (FNA); Fig. 5R shows facial descriptor measurement Nasal Spine Deviation Proportion (NSDP); Fig. 5S shows facial descriptor measurement Nasal Tip Inflection Proportion (NTIP); Fig. 5T shows facial descriptor measurement Alar Flare Proportion (AFP); Fig. 5U shows facial descriptor measurement Nasal Bridge Width Proportion (NBWP); 25 Fig. 5V shows facial descriptor measurement Nasal Spine Angle (NSA); Fig. 6A shows facial descriptor measurement Brow Fullness Proportion (BFP); Fig. 6B shows facial descriptor measurement Forehead Inflexion Proportion (FIP); Fig. 6C shows facial descriptor measurement Forehead Slope Proportion (FSP); Fig. 6D shows facial descriptor measurement Forehead Roundness Proportion (FRP); 30 Fig. 6E shows facial descriptor measurement Forehead Maxima Point Proportion

(FMPP);

WO 2013/147932 PCT/US2012/064178 9 Fig. 6F shows facial descriptor measurement Forehead Height Proportion (FHP); Fig. 6G shows facial descriptor measurement Eyebrow Height Proportion (EHP); Fig. 6H shows facial descriptor measurement Forehead Width Proportion (FWP); Fig. 6J shows facial descriptor measurement Forehead Fullness Proportion (FFP); 5 Fig. 6K shows facial descriptor measurement Hairline Inflection Angle (HIP); Fig. 7A shows facial descriptor measurement Auricle Inclination Angle (AIA); Fig. 7B shows facial descriptor measurement Auricle Set Angle (ASA); Fig. 7C shows facial descriptor measurement Ear-to-Eyebrow Height Proportion (EtEHP); 10 Fig. 7D shows facial descriptor measurement Ear-to-Eye Height Proportion (EtEHP2); Fig. 7E shows facial descriptor measurement Lobe Attachment Proportion (LAP); Fig. 7F shows facial descriptor measurement Lobe Length Proportion (LLP); Fig. 7G shows facial descriptor measurement Intertragical Notch Depth Proportion (INDP); 15 Fig. 7H shows facial descriptor measurement Tragus Protuberance Proportion (TPP); Fig. 7J shows facial descriptor measurement Tragus Mid-Point Proportion (TMPP); Fig. 7K shows facial descriptor measurement Antitragus Protuberance Proportion (APP); Fig. 7L shows facial descriptor measurement Antitragus Mid-Point Proportion (AMPP); Fig. 7M shows facial descriptor measurement Concha Inflexion Proportion (CIP); 20 Fig. 7N shows facial descriptor measurement Upper Concha Separation Angle (UCSA); Fig. 7P shows facial descriptor measurement Ear Tip Length Proportion (ETLP); Fig. 7Q shows facial descriptor measurement Upper Concha Separation Angle (UCSA); Fig. 7R shows facial descriptor measurement Ear Roundness Proportion (ERP); Fig. 7S shows facial descriptor measurement Crus Helix Angle (CHA); 25 Fig. 7T shows facial descriptor measurement Tubercle Proportion (TP); Fig. 7U shows facial descriptor measurement Helix Thickness Proportion (HTP); Fig. 7V shows facial descriptor measurement Lobule Inflexion Proportion (LIP); Fig. 8 shows various relationships between genes, hormones, behavior, and facial features of an individual, such as a human or other animal; 30 Fig. 9 shows a system that can be used to implement any of the methods of the present disclosure; WO 2013/147932 PCT/US2012/064178 10 Fig. 10 shows an exemplary receiver operating characteristic (ROC) curve; Fig. 11 shows a method for determining a predictor of a characteristic of an individual according to an embodiment of the present disclosure; Fig. 12 shows a method for determining a characteristic of an individual according to 5 another embodiment of the present disclosure; Fig. 13 shows another method for determining one or more predictors of a characteristic of an individual, according to other embodiments of the present disclosure; Fig. 14 shows another method for determining a characteristic of an individual according to other embodiments of the present disclosure; 10 Fig. 15 shows an example illustrating an embodiment of the method shown in Fig. 14; Fig. 16 is a flowchart of a method for determining a predictor for the pairwise complementarity of two individuals for a particular purpose, according to other embodiments of the present disclosure; Fig. 17 is a flowchart of a method for predicting a characteristic of a pair of individuals, 15 such as pairwise complementarity for a particular purpose, according to other embodiments of the present disclosure; Fig. 18 is a flowchart of an exemplary method for determining the suitability of a new individual as a member of an existing group based on the predicted conflict with the existing group members, according to other embodiments of the present disclosure; and 20 Fig. 19 shows a method for determining a strategy for a group of individuals using facial descriptor measurements, according to another embodiment of the present disclosure. DETAILED DESCRIPTION [0014] Various methods for using metrics, such as facial measurements, associated with the physical form of an individual, such as a human or other animal, to predict a 25 characteristic including behavior, suitability, or potential are described. Also, the methods described are useful for grouping, pairing, or matching of individuals for various purposes, both within a species and across species. [0015] The biological mechanism relied upon in the disclosed embodiments was first proposed in the ground-breaking 1999 Russian study entitled "Early Canid 30 Domestication: The Fox Farm Experiment." Using an extensive breeding program of WO 2013/147932 PCT/US2012/064178 11 wild silver foxes (vulpes vulpes), this study showed that selective breeding can be used to alter the innate personality traits or cognitive characteristics of a line of domesticated animals. More particularly, this study demonstrated that endocrine or hormone changes are significantly correlated with both the personality trait changes and a predictable suite 5 of morphological changes, most predominantly in the structures of the face. [0016] As illustrated graphically in Fig. 8, the innate personality of an animal originates in its genetic composition. Genes dictate the basal levels of neurologically active hormones that control the behavior of an animal, such serotonin, which inhibits aggression. Genes also dictate the timing of the release of these hormones, such as 10 corticosteroids (stress response) that control the windows of postnatal cognitive and social development in young animals. The cognitive framework of an animal is thus determined from a combination of these innate personality traits provided by this genetically controlled endocrine makeup - the so-called "nature effect" - and the stimuli and experiences that the animal was subject to during development - the so-called 15 "nurture effect." When viewed in the context of the animal's current environment, this cognitive framework dictates its behavioral performance, which may be defined in such terms as cognitive suitability to a specific task, success in performing a specific task, likelihood of displaying a specific type or pattern of behavioral responses, or, when compared against the personality types of its conspecifics, performance of an individual 20 in group situations. [0017] As mentioned above, variation in the basal levels of neurologically active hormones and in particular the timing of their release windows during development account not only for differences in innate personality among animals of the same species, but also for variability in morphology, particularly of the face. This direct correlation 25 between the facial structure and endocrine composition of an animal subsequently allows for quantifiable features of an animal's face to be correlated with and used as a proxy for predicting variability in the innate behavior of individual animals as a function of their neurochemical makeup. Variations in facial structure may also be used to predict the behavior and performance of an animal as a result of the variations in the degree of 30 functionality that they allow, in terms such as field of vision, auditory acquisition, oxygen intake, feed intake, etc.

WO 2013/147932 PCT/US2012/064178 12 [0018] Various facial recognition and image matching techniques will mathematically model an animal's face and allow the prediction of behavior and performance. While these embodiments are effective, the processes and techniques of facial recognition and image matching are generally computationally intensive. Therefore, trigonometric 5 modeling is used by some embodiments. Combinations of facial/shape recognition, image matching and trigonometric modeling may be used to predict behavior and performance. [0019] Since the domestication phenotype is common across many species of animals, embodiments described herein are applicable to a broad range of individuals, both 10 mammals and non-mammals, including humans, donkeys, cattle, oxen, llamas, sheep, goats, turkey, geese, dogs, foxes, cats, ferrets, camels, geese, chickens, pigs, fish, etc. For example, embodiments described herein may be used to predict certain characteristics of humans via methods employing facial descriptor measurements. During progressive stages of human embryonic growth, development of the brain and face remains intimately 15 connected through both genetic signaling and biomechanical/biochemical mechanisms. The face develops from populations of cells originating from the early neural crest, with cells from the neural tube gradually shifting to form the prominences of the face. Differentiation of these early cells is closely regulated through intricate genetic signaling mechanisms, with the brain essentially serving as the platform on which the face grows. 20 As these two structures continue to grow and develop during the later embryonic stages, their phenotypes remain closely linked as complex genetic hierarchies regulate patterns of cross talk between molecules, cells, and tissues. [0020] For example, the close relationship between the functional development of the brain and structures of the face has been clearly documented for a number of 25 developmental disorders. Among the most well known of these disorders is Fetal Alcohol Syndrome (FAS), which is the direct result of exposure of the fetus to alcohol during pregnancy. FAS has been shown to result in both an easily identifiable phenotype (i.e., collection of minor facial abnormalities such as small eyes, smooth philtrum, thin upper lip) and developmental damage to the central nervous system that is often permanent 30 (e.g., speech delays, learning disabilities, poor reasoning skills, poor memory, attention deficit disorders, and low IQ). Figs. 1 through 7 show a set of exemplary human facial WO 2013/147932 PCT/US2012/064178 13 descriptor measurements that can be used to identify the phenotype associated with FAS. For example, Figs. 1A through 1X show various facial descriptor measurements related to the eye, while Figs. 3A through 3U show various facial descriptor measurements related to the mouth. 5 [0021] By way of further example, Down Syndrome is another prenatal developmental disorder causing mental/social delays that yields an easily identifiable phenotype including a host of distinctive facial features such as small chin and/or mouth, round face, and rounded almond-shaped eyes. Recent studies have even been able to identify measurable trends in facial features that distinguish between children diagnosed with 10 Autism Spectrum Disorders (ASD) and those of typical development. The facial descriptor measurements shown in Figs. 1 through 7 may also be used to identify the phenotype associated with Down syndrome. However, as with FAS described above, this set of facial descriptor measurements is merely exemplary, and may be reduced or augmented for predicting Down syndrome or other human functional development 15 disorders within the spirit and scope of the present disclosure. [0022] The set of human facial descriptor measurements shown in Figs. 1 through 7 are merely exemplary, and the person of ordinary skill will recognize that fewer than the entire set may be used to predict FAS or Down Syndrome. Furthermore, this set of human facial descriptor measurements is not exhaustive and others may be incorporated. 20 In some embodiments, two, three, or more of the facial descriptor measurements are used in combination to predict a trait, characteristic, or syndrome such as FAS or Down Syndrome. Moreover, while examples are shown primarily with facial measurements, other head measurements and physical measurements may be used alone, without facial measurements, or in combination with facial measurements. For example, measurements 25 of the head or crown may be used in conjunction with facial features to predict syndromes or traits. [0023] Given these established relationships between human facial structures and cognitive development, any of the computationally inexpensive, two-dimensional, locally-normalized facial evaluation systems described provides a non-invasive analysis 30 tool for use in multiple clinical applications. For example, embodiments of the facial analysis methods and systems disclosed herein will diagnose children with specific WO 2013/147932 PCT/US2012/064178 14 cognitive development disorders based on the degree of divergence between their facial features and those of the overall typical population with respect to the phenotype for a disorder. Such a diagnosis tool is faster and less invasive than the standard cognitive testing procedures currently employed, and may allow for earlier diagnosis and 5 interventions. More computationally expensive embodiments or variations may also be used for diagnosis. [0024] More generally, embodiments of the facial analysis methods and systems disclosed herein also can be used for predicting or inferring how an individual is innately programed to perceive, synthesize, and then respond to different types of stimuli. Facial 10 analysis can be applied to predict not only how the cognitive framework of an individual will interact with and be shaped by their environment, but also how it will interact with the cognitive frameworks of other human beings, as described in more detail below. [0025] Each individual of a species (including humans) experiences the world differently due to the unique way in which the individual perceives the same set of 15 stimuli. For example, pain is a stimulus whose perception varies widely among individuals. Although cultural and social expectations arguably influence how readily an individual admits to discomfort, it is clear that some people are more sensitive to pain than others, due in large part to greater capability to block out and cope with pain. Although some of this variability is due to the structure and number of neurons in the 20 peripheral nervous system that register a pain input, how easily this input is brought to the forefront of a person's consciousness also plays a major role in overall pain sensitivity. As such, the cognitive structures for overcoming and dealing with such unpleasant sensations play a major part in determining an individual's resistance to pain. As a result, facial analysis methods can be applied to develop a useful model or tool that 25 objectively predicts an individual's perception of the severity of pain. [0026] Compared to the subjective pain assessments methods currently in use (e.g., scale of 1 to 10), such a model would benefit the medical community in many different ways. In addition simply helping doctors more efficiently and correctly assess the severity of an injury on their initial diagnosis, such a tool could also be used to more 30 fairly assign priority to patients in waiting room settings. An objective pain assessment model also would be useful in preventative medicine and recovery care, particularly in the WO 2013/147932 PCT/US2012/064178 15 case of patients with high pain thresholds, who in underscoring the severity of their condition might otherwise keep doctors from readily picking up on health concerns in its earlier and more treatable stages. In addition, such a model could also be used in helping doctors to more rigorously qualify the strength and quantity of pain medications assigned 5 to patients after a procedure. For example, if information from such an objective model indicates that a patient has a lower-than-average pain tolerance, a physician may decide to prescribe a higher strength pain medication to ensure that the patient will be able to rest comfortably enough to recover fully from their injury and to cope with the discomfort of their physical therapy sufficiently enough to regain full mobility. Likewise, this objective 10 model might also indicate that a patient with a notably high pain threshold would do just fine with a lower level pain medication following a procedure that people with a normal pain threshold would otherwise find to simply have too difficult of a recovery period to cope with without the support of stronger medications. In this way, doctors could tailor the amount of pain medication prescribed based on the individual patient's objective pain 15 perception, reducing the unnecessary exposure of the patient and the overall distribution and subsequent unauthorized or illegal access to these highly addictive medications. [0027] Another application of the facial analysis-based tool for objectively predicting an individual's pain perception is to help orthopedists determine when surgery is an appropriate option. Major surgeries to joints such as the knees, hips, and shoulders are 20 typically followed by a recovery period during which most patients undergo a regular schedule of painful physical therapy in order to regain full range of motion of the joint. Failure to follow this post-surgical therapy regime will result in permanent stiffening and tightening of the joint, even to the point of complete immobilization - a so called "frozen joint." Due to such risks, most orthopedists try to avoid major surgeries as long as 25 possible, typically until and even after chronic pain has begun to seriously impact the patient's quality of life. By using the information gleaned from facial inference technologies about their patients' innate tolerance of or sensitivity to pain, orthopedists can more accurately predict how well a patient will be able to follow the post-surgical therapy and, thus, the likelihood of a full recovery. With this ability, orthopedists can 30 make informed decisions tailored to the individual patient about if and when the benefits outweigh the risk to pursue surgical options.

WO 2013/147932 PCT/US2012/064178 16 [0028] Another group of applications for the facial analysis methods described herein relate to predicting an individual's sensitivity to social cues. Individual humans display a wide range of sensitivity in their ability to readily pick up on and correctly perceive social cues. On the most sensitive side of this spectrum are Highly Sensitive Persons (HSPs), 5 who innately pick up on even the subtlest of social cues. Such individuals, who are thought to make up roughly 15% of the population, are acutely and at times even painfully aware of the thoughts and feelings of others during social situations, and are subsequently extremely sensitive to the reactions of others. While they can be extremely responsive to praise, they also tend to take even the lightest criticisms and reprimands 10 harshly. As a result, they are more heavily influenced and reflective of social experiences than most people, particularly during their early childhood, and are more susceptible to being shy, becoming introverted, and performing poorly at tasks when they feel that they are under scrutiny. [0029] On the opposite end of this spectrum are individuals with Autism Spectrum 15 Disorders (ASDs), who show a very low sensitivity to perceiving the social cues of others. In some ASDs, such as Asperger's Syndrome, individuals are unable to properly synthesize and correlate different facial expressions and tones of voice to the correct emotional state of others. In all ASDs, however, it is readily observed and understood that such individuals simply do not pick up on social cues as readily as the general 20 population, typically because they are too distracted by extraneous environmental details to notice cues from their social environment. As a result, attempting to communicate with individuals suffering from ASDs can be fairly frustrating for others who are unaccustomed to or lack the patience to deal with this lack of focus and responsiveness. Individuals with ASDs, including those who are even deemed high functioning, also tend 25 to have a difficult time maintaining interpersonal relationships due to their inability to connect and relate with what others are feeling. [0030] Distinct patterns in the facial features that distinguish individuals with ASDs from those in the general population have long been noted by the medical and psychiatric communities. Moreover, recent research has shown that computer-generated facial 30 measures can identify and differentiate between children with such types of conditions. Using the facial analysis methods described herein, similar models can be used to identify WO 2013/147932 PCT/US2012/064178 17 HSPs and those suffering from ASDs from among the general population. In addition to identifying and classifying persons in these outlying groups, these facial analysis methods can be used to model the social sensitivity of the general population on a continuous scale, and to predict the social sensitivity of any one individual. 5 [0031] By further example, the facial analysis methods described herein also can be applied to model and predict how individuals respond to aggression. In general, the range of innate responses to aggression varies widely within the general population. Some individuals are by nature very easily offended, and tend to take the statements and actions of others more personally than intended. Such individuals subsequently tend to be much 10 more bellicose in nature, and more readily engage in confrontation with others. On the other hand, some individuals regard very few statements and actions of others as offensive, even when they were meant to be. These individuals are frequently seen as having a "long fuse" or low temper. The ability to objectively model and predict an individual's relative perception of aggressive behavior of others has many benefits. For 15 example, such information would help determine the suitability of an individual to certain tasks or jobs, particularly those that require considerable interaction with customers, working in a culturally diverse team environment, or dealing with situations where aggressive behavior is the norm (e.g., law enforcement). Also, such a model could also prove valuable in dealing with criminals themselves, helping law enforcers to identify 20 individuals that pose a greater risk of getting into physical confrontations either within the general public or within a prison setting. [0032] For example, a model based on a plurality of facial measurements may be used in the manner described below with reference to Figs. 11 through 14 to predict a particular individual's innate personality such as aggression and competitiveness. One or 25 more of the facial descriptor measurements shown in Figs. 1 through 7 may also be used for this purpose. However, this set of facial descriptor measurements is merely exemplary and may be reduced or augmented as necessary within the spirit and scope of the present disclosure. Furthermore, multivariate analysis of a plurality of facial measurements statistically normalized against like measurements of a standard population 30 of individuals can be used to predict other variable aspects of innate personality such as aggression, dominance, conscientiousness, compassion, extroversion, IQ, etc. Insight WO 2013/147932 PCT/US2012/064178 18 into such personality traits of an individual could be used to predict various aspects of behavior and performance such as learning styles, athletic performance, business decisions, etc. [0033] For example, each year Major League Baseball teams attempt to find the best 5 young players to select in the amateur draft. Although the draft contains multiple rounds, the first round is especially important because this is where the best players are chosen. Each of these first-round draft picks are typically compensated by a multi-million dollar signing bonus. Given their relative rarity, outstanding left-handed pitchers often are the most highly sought-after first-round draft picks, and often command larger signing 10 bonuses that other players. However, many of these pitchers are relatively young and unproven. The cognitive suitability of a pitcher to performing under the extreme pressure situations found in Major League Baseball is a critical component for success that is otherwise very difficult to measure using conventional tools currently available to baseball scouts. Although baseball scouts use many different techniques (e.g., radar 15 guns) to measure pitchers' physical capabilities, often these players' mental and cognitive capabilities - so-called "intangibles" -remain relatively unknown at draft time. Accordingly, it would be advantageous for teams to have a reliable indicator of the mental and cognitive capabilities of prospective first-round draft picks, such as left handed pitchers, prior to making a decision to invest a large signing bonus and take the 20 opportunity cost of foregoing another top prospect. [0034] Facial and other physical measurements of successful left-handed baseball pitching prospects (or any other baseball prospects) can be made and compared with measurements of a standard population. Digital images and reference points may be used to ease the burden of the measurement process. Ratios, angles, or nominal measurements 25 may be used instead of or in addition to actual measurements. A number of metrics may then be analyzed to find those metrics which show the greatest correlation or significance in determining statistically whether a person is more likely than another to be a successful left-handed pitcher. Once the more significant metrics are identified, the system may simply rely on those metrics to predict a person's likelihood to succeed as a left-handed 30 pitcher. A baseball scout may use the system as a tool to assist in choosing one left handed pitching prospect over another by scoring both candidates. Although the WO 2013/147932 PCT/US2012/064178 19 advantages have been illustrated by the example of a left-handed pitcher, those skilled in the art will recognize that such advantages could be achieved by use of the same or similar methods with respect to other baseball players or players in other sports using a draft, such as the National Football League (NFL) and the National Basketball 5 Association (NBA). Moreover, those skilled in the art will recognize numerous other applications of the methods and techniques described throughout the specification to humans. [0035] Fig. 11 shows an embodiment in the form of a method for determining a predictor of a characteristic of a particular type of individual (e.g., a human). As used 10 herein, "type" refers to breed, species, sub-species, or any other relevant genetic similarity. Also, "characteristic" is used broadly to refer to a wide variety of features, traits, behaviors, capabilities, suitability, etc., including the most basic or fundamental personality features (e.g., sensitivity); more complex personality features that may be comprised of multiple basic features; and overall suitability for a particular task, purpose, 15 relationship, etc., which may be influenced by a variety of fundamental and complex personality features. Accordingly, given the appropriate inputs, the method of Fig. 11 can be used to predict any type or complexity of characteristic. Although Fig. 11 illustrates the one or more embodiments by blocks arranged in a specific order, this order is merely exemplary and the steps or operations comprising the method may be performed in a 20 different order than shown in the figure. Moreover, a person of ordinary skill will understand that the blocks shown in Fig. 11 may be combined and/or divided into blocks having different functionality. [0036] In block 1100, a sample library is created. This step comprises obtaining one or more digital images of an anatomical region for each of a plurality of individuals, e.g., the 25 face of a horse or a human. The image may be obtained in various ways, such as from a memory card of a digital camera, by downloading via File Transfer Protocol (FTP), via email, etc. Once obtained, the images are stored in a memory operably connected to a digital computer, such as memory that is either locally or remotely accessible by the computer, including an attached hard drive, removable flash drive, network drive, RAID 30 (redundant array of independent disks) drive, removable memory card, etc.

WO 2013/147932 PCT/US2012/064178 20 [0037] Block 1100 also comprises obtaining additional data related to the characteristic of interest and storing it in the same or a different memory operably connected to the digital computer. As used herein, "data" may refer to performance records, vital statistics, results of behavior studies, and/or any other quantitative information that is 5 related in some way to the characteristic of interest. While qualitative data may also be used, in many embodiments the qualitative data is converted into quantitative data for easier use by the system. The additional data may pertain to individuals in the sample library, groups of individuals in the sample library, or more generally the type of individual. Additional data may comprise information on breeding (e.g., 10 pedigree/ancestry), health, and environmental factors (e.g., training, experiences). The additional data is stored in a manner and location by which it can be associated with the sample library images, such as in a relational database accessible by the digital computer. [0038] In block 1102, a plurality of reference points are added to the one or more stored images of a particular individual in the sample library. This may be accomplished in an 15 automated fashion or manually, for example by use of a program with a graphical user interface (GUI) running on the digital computer. For example, the one or more images can be processed by MATLAB, an advanced mathematical analysis program sold by The Mathworks, Inc., of Natick, Massachusetts (http://www.mathworks.com). MATLAB provides advanced image processing features and flexible options for definition of large 20 numbers of variables, specifically matrices. Reference points are added to each of the images by using the MATLAB "GInput" command, which provides an interactive selection GUI. In some embodiments, reference points are manually selected on the image. One such embodiment is shown in Figs. 1 through 7, which were manually annotated with reference points (e.g., the four points in Fig. 1A). In other embodiments, 25 reference points may be added automatically by MATLAB or another software application based on a generic model of the individual's face. Once the reference points are entered onto an image, MATLAB maps their pixel locations within the image to numerical coordinates within the corresponding matrix to facilitate further computation. [0039] In block 1104, one or more facial descriptor measurements (FDMs) related to 30 the characteristic of interest are computed from the set of reference points that were added to the one or more digital images of the individual. The facial descriptor measurements WO 2013/147932 PCT/US2012/064178 21 may be computed using distance measurements and trigonometric functions as illustrated above. Because length distance measurements are based on coordinate positions within the pixel matrix, the absolute distance values may be sensitive to factors such as camera resolution, artifacts of one or more compressions of the image, and cropping applied to 5 isolate the face. In some embodiments, the length measurements may be normalized to structural reference lengths that are effectively constant among individuals of the same type and subject to the same set of factors. In other embodiments, the length measurements may be normalized against measurements (e.g., length or area) of other features in the same area or region of the face. For example, a measure of the lip 10 thickness may be normalized by the lip length or width. However, it is apparent to one of ordinary skill that the facial descriptor measurements may be based upon absolute or non normalized length measurements if the factors described above were not present or were not a concern. In other embodiments, one or more of the facial descriptor measurements may be based on an angular measurement or an area measurement. The facial descriptor 15 measurements may be based on non-trigonometric calculations, such integral calculus calculations, or a combination of trigonometric and non-trigonometric calculations. [0040] In some embodiments, one or more of the digital images are three-dimensional images. Such images may be created by combining multiple two-dimensional images using stereophotogrammetry or other methods known to persons of ordinary skill in the 20 art. In such embodiments, one or more of the facial descriptor measurements may be based on a three-dimensional measurement, such as an absolute distance measure, absolute volume, a volumetric ratio, a solid angle, a dihedral angle, a surface area, or a combination thereof. [0041] As shown in Fig. 11, blocks 1102 and 1104 are repeated for each individual in 25 the sample library. Once complete, in block 1108, a relationship is determined between a particular facial descriptor measurement and the additional data related to the characteristic of interest using all individuals in the sample library. For example, the relationship can be determined from the mathematical correlation between the facial descriptor measurement and additional data for all individuals in the sample library. The 30 correlation may be normalized or scaled as necessary to make it meaningful for further computation or interpretation.

WO 2013/147932 PCT/US2012/064178 22 [0042] Other measures can be used to determine the relationship of a facial descriptor measurement to the additional data. For example, in categorical models (i.e., those used to separate individuals into categories and/or subcategories), receiver operating curve (ROC) analysis may be used to determine how effectively a facial descriptor 5 measurement categorizes the collection of individuals in the sample library according to the characteristic of interest. In other words, if the additional information associated with the images in the sample library includes the actual categories and/or subcategories that the individuals fall into (e.g., personality type/subtype, consumption/purchasing patterns, learning style, type of event, etc.), ROC analysis can be used to determine how well a 10 facial descriptor measurement assigns individuals to their actual category (i.e., true positive rate) while avoiding assignment of individuals to incorrect categories (i.e., false positive rate). Fig. 10 shows an exemplary receiver operating curve, where the false positive rate is shown on the horizontal axis and the true positive rate is shown on the vertical axis. The curve shown in Fig. 10 has an area of 0.801; an area greater than 0.65 15 indicates that a particular facial descriptor measurement is useful for prediction or categorization. Alternately, multi-dimensional, Euclidean distance analysis also may be used to separate two groups categorically. Other methods for determining a relationship based on appropriate statistical models are apparent to persons of ordinary skill in the art. As illustrated in Fig. 11, block 1106 is repeated for each facial descriptor measurement. 20 [0043] In block 1112, one or more of the facial descriptor measurements are selected to be used as predictors of the characteristic of interest. Any number of facial descriptor measurements - up to and including the entire set - may be selected. If there are multiple characteristics of interest, then a separate selection may be made for each characteristic. The selection may be made in various ways, depending on the available information and 25 the characteristic of interest. The selection may be based on the raw facial descriptor measurements themselves, or upon scaled or transformed versions. For example, the z score of a particular facial descriptor measurement may be used for purposes of selection. As known to persons of ordinary skill, a z-score represents the relative frequency (or rarity, as the case may be) of a particular measurement within a population, and is also 30 commonly referred to as a standard score or z-value. Other transformations will be apparent to persons of ordinary skill in the art.

WO 2013/147932 PCT/US2012/064178 23 [0044] Moreover, in block 1112, the combination of the selected facial descriptor measurements that optimizes the predictor is also determined. In some embodiments, an optimal linear combination of the selected subset of facial descriptor measurements is determined using statistical analysis techniques. For example, in categorical models, 5 ROC analysis may be used to select which combination of facial descriptor measurements best categorizes the collection of individuals in the sample library according to the characteristic of interest. As described above with respect to block 1108, ROC analysis can be used to determine which combination of facial descriptor measurements best assigns individuals to their actual category (i.e., true positive rate) while avoiding 10 assignment of individuals to incorrect categories (i.e., false positive rate). In the spirit of the disclosure, however, a non-linear combination of the entire set, or a selected subset, of the facial descriptor measurements also may be determined from the sample library. A non-linear combination of the selected facial descriptor measurements may be selected using optimization techniques including, but not limited to, Newton's Method and 15 Lagrange's Method. Moreover, the selected linear and/or non-linear combination may be of the facial descriptor measurements themselves, or of scaled or transformed versions such as the z-scores. If multiple characteristics are of interest, then an optimal combination for each characteristic may be selected. [0045] Fig. 13 is a flowchart of an alternative embodiment of a method for determining 20 a predictor of a characteristic of a particular type of individual (e.g., a human). Although Fig. 13 illustrates the one or more embodiments by blocks arranged in a specific order, this order is merely exemplary and the steps or operations comprising the method may be performed in a different order than shown in the figure. Moreover, a person of ordinary skill will understand that the blocks shown in Fig. 13 may be combined and/or divided 25 into blocks having different functionality. [0046] The method of Fig. 13 explicitly recognizes that more complex characteristics are based on combinations of more fundamental characteristics and overall suitability for a particular task, purpose, relationship, etc., may be based on a variety of fundamental and complex characteristics. Accordingly, the method of Fig. 13 can determine a model 30 comprised of predictors of basic characteristics (called "principal indicators"), which are based on facial descriptor measurements; predictors of more complex characteristics WO 2013/147932 PCT/US2012/064178 24 (called "complex indicators"), which may be based on, or combinations of, facial descriptor measurements and/or principal indicators; and predictors of an ultimate suitability or category for an individual (called "suitability indicators"), which may be based on any of the facial descriptor measurements, principal indicators, and complex 5 indicators. Although not shown in Fig. 13, persons of ordinary skill will readily comprehend that the model may comprise one or more additional levels of indicators. For example, the model may comprise one or more "tertiary indicators" that may be based on, or combinations of, facial descriptor measurements, principal indicators, and/or complex indicators. In such case, the one or more suitability indicators may be based on 10 any of the facial descriptor measurements, principal indicators, complex indicators, and tertiary indicators. Moreover, models comprising such additional levels of indicators may be used in other methods described herein for predicting characteristics of one or more individuals, such as shown in and described below with reference to Fig. 14. [0047] In block 1300, a sample library comprised of data for N individuals or subjects 15 is created. This operation comprises obtaining one or more digital images of an anatomical region for each of the subjects, e.g., a human face. As discussed above with reference to Fig. 11, the images may be two- or three-dimensional images. The images may be obtained in various ways, such as from a removable or non-removable memory of a camera device (e.g., digital camera, webcam, smartphone, tablet, etc.), by downloading 20 via File Transfer Protocol (FTP); by streaming from a camera device, via email, etc. Once obtained, the images are stored in a memory operably connected to a digital computer, such as memory that is either locally or remotely accessible by the computer, including an attached hard drive, removable flash drive, network drive, RAID (redundant array of independent disks) drive, removable memory card, etc. 25 [0048] Block 1300 also comprises obtaining additional data related to the subjects and storing it in the same or a different memory operably connected to the digital computer. The additional data may relate specifically to the individuals in the sample library, groups of individuals in the sample library, or more generally to a population of which the individuals are a part. As used herein, "additional data" may refer to performance 30 records, vital statistics, results of behavior studies, environment data, and/or any other quantitative information that is related in some way to the characteristic of interest.

WO 2013/147932 PCT/US2012/064178 25 While qualitative data may also be used, in many embodiments the qualitative data is converted into quantitative data for easier use by the system. Moreover, additional data may be comprised of principal additional data, which are related to the principal indicators of interest; complex additional data, which are related to the complex 5 indicators of interest; and categorical additional data, which are related to the ultimate suitability indicator or categorization. The additional data is stored in a manner and location by which it can be associated with the sample library images, such as in a relational database accessible by the digital computer. [0049] In block 1302, the images corresponding to each of the N individuals or subjects 10 in the sample library are annotated with a plurality of reference points. As discussed above with reference to Fig. 11, the annotation may be done manually or automatically. Also in block 1302, a set of M facial descriptor measurements (Fi, i = 1 to M) are computed based on the reference points annotated onto the one or more images for each of the N individuals. The facial descriptor measurements Fi may be computed using 15 absolute or normalized distance measurements, angular measurements, area measurements, volume measurements, etc. using trigonometric functions, integral calculus, and other computational methods known to persons of ordinary skill. Although block 1302 shows both the annotation and the computation of Fi being done sequentially for each subject, the person of ordinary skill will recognize that these operations could be 20 arranged differently, e.g., annotations for all subjects followed by computation of Fi for all subjects. [0050] In block 1304, one or more relationships is determined between each of the facial descriptor measurements F;, i = 1 to M, and the principal additional data across all N subjects in the sample library. By way of example, the one or more relationships for 25 facial descriptor measurements Fi can be determined using one or more mathematical correlations between Fi and at least some portion of the principal additional data for all N subjects in the library. The correlation may be normalized, scaled, or transformed (e.g., z-score) as necessary to make it meaningful for further computation or interpretation. As discussed above with reference to Figs. 11 and 12, in some embodiments, receiver 30 operating curve (ROC) analysis may be used to determine the effectiveness of facial descriptor measurement Fi as a principal indicator, based on the ability of Fi (or a function WO 2013/147932 PCT/US2012/064178 26 thereof) to assign individuals to their actual category (i.e., true positive rate) while avoiding assignment of individuals to incorrect categories (i.e., false positive rate). In other embodiments, correlation analysis may be used to determine the one or more relationships. Other methods for determining a relationship based on appropriate 5 statistical models are apparent to persons of ordinary skill in the art. [0051] In block 1306, it is determined whether principal indicators Pk, k = 1 to K, are to be used in the computation of the suitability indicator, S. In the case that only facial descriptor measurements Fi, i = 1 to M, are to be used in computing S, the method proceeds to block 1318. Otherwise, the method proceeds to block 1308, where one or 10 more of the facial descriptor measurements, Fi, are selected to be used to compute each of principal indicators Pk, k = 1 to K. Any number of facial descriptor measurements - up to and including the entire set - may be selected for each principal indicator. The selection may be made in various ways, depending on the available information and the particular principal indicator, Pk. The selection may be based on the raw facial descriptor 15 measurements themselves, or upon scaled or transformed versions. For example, the z score of a particular facial descriptor measurement may be used for purposes of selection. Other transformations will be apparent to persons of ordinary skill in the art. [0052] In block 1308, the combination of the selected facial descriptor measurements that optimizes each of principal indicators Pk is also determined. In some embodiments, 20 an optimal linear combination of the selected subset of facial descriptor measurements is determined. In other words, if Pk = a1Fkl + a2Fk2 + ... anFkn, then the facial descriptor measurements Fki, i = 1 to n, and the linear coefficients ai, i = 1 to n, are determined in block 1308. The linear combination may be determined using statistical analysis techniques. For example, in categorical models, ROC analysis may be used to select the 25 combination of facial descriptor measurements that provides a Pk that best categorizes the subjects in the sample library, e.g., according to at least a portion of the principal additional data. In the spirit of the disclosure, however, a non-linear combination of the entire set, or a selected subset, of the facial descriptor measurements also may be determined based on the principal additional data. A non-linear combination of the 30 selected facial descriptor measurements may be selected using optimization techniques including, but not limited to, Newton's Method and Lagrange's Method. Moreover, the WO 2013/147932 PCT/US2012/064178 27 selected linear and/or non-linear combination may be of the facial descriptor measurements themselves, or of scaled or transformed versions such as the z-scores. [0053] In block 1310, it is determined whether complex indicators C;, j = 1 to J, are to be used in the computation of the ultimate suitability indicator, S. If not, then the method 5 proceeds to block 1318. Otherwise, the method proceeds to block 1312, where one or more relationships are determined between each of the principal indicators Pk, k = 1 to K, and the complex additional data across all N subjects in the sample library. By way of example, the one or more relationships for principal indicators Pk can be determined using one or more mathematical correlations between Pk and the complex additional data for all 10 N subjects in the library. The correlation may be normalized, scaled, or transformed (e.g., z-score) as necessary to make it meaningful for further computation or interpretation. As discussed above, in some embodiments, the relationships can be determined based on receiver operating curve (ROC) analysis. In other embodiments, correlation analysis may be used to select the complex indicators used for computing the ultimate suitability 15 indicator, S. In block 1312, one or more relationships are also determined between each of the facial descriptor measurements F;, i = 1 to M, and the complex additional data across all N subjects in the sample library. ROC analysis, correlations, or other methods known to persons of ordinary skill may be used to determine these relationships. [0054] Next, in block 1314, one or more of the facial descriptor measurements, Fi, and 20 principal indicators, Pk, are selected to be used to compute each of complex indicators C;, j = 1 to J. Any number of facial descriptor measurements and principal indicators - up to and including the entire set of both - may be selected for each principal indicator. Moreover, the selection may be limited only to one or more of the principal indicators, Pk, i.e., the facial descriptor measurements are not used. The selection may be made in 25 various ways, depending on the available information and the particular complex indicator, C. The selection may be based on the raw values of Fi and Pk, or upon scaled or transformed versions. For example, the z-score of a particular F; or Pk may be used for purposes of selection. Other transformations will be apparent to persons of ordinary skill in the art. 30 [0055] In block 1314, the combination of the selected facial descriptor measurements, Fi, and/or principal indicators, Pk, that optimizes each of complex indicators Cj is also WO 2013/147932 PCT/US2012/064178 28 determined. In some embodiments, an optimal linear combination of the selected group of Fi and/or Pk is determined. In other words, if Cj = a1Fki + a 2Fk2 + ... a nF + p1Pk1 + p2Pk2 + ... PmPmkn, then the facial descriptor measurements Fki, i = 1 to n, and principal indicators Pki, i = 1 to m, are selected in block 1312 and the linear coefficients a ki, i = I to 5 n, and P ki, i = 1 to m, are determined in block 1314. The linear combination may be determined using statistical analysis techniques. In some embodiments, ROC analysis may be used to select the combination of facial descriptor measurements and/or principal indicators that provides a C that best categorizes the subjects in the sample library, e.g., according to at least a portion of the complex additional data. In other embodiments, 10 correlation analysis may be used to select the combination of measurements. [0056] In the spirit of the disclosure, however, a non-linear combination of the selected group of facial descriptor measurements and/or principal indicators also may be determined based on the complex additional data. A non-linear combination of the selected group of facial descriptor measurements and/or principal indicators may be 15 determined using optimization techniques including, but not limited to, Newton's Method and Lagrange's Method. Moreover, the selected linear and/or non-linear combination may be of the facial descriptor measurements and/or principal indicators themselves, or of scaled or transformed versions such as the z-scores. [0057] In block 1314, one or more relationships are also determined between each of 20 the complex indicators C; and the categorical additional data across all N subjects in the sample library. By way of example, the one or more relationships for complex indicators Cj can be determined using one or more mathematical correlations between C; and the categorical additional data for all N subjects in the library. The correlation may be normalized, scaled, or transformed (e.g., z-score) as necessary to make it meaningful for 25 further computation or interpretation. As discussed above, in some embodiments the relationships can be determined based on receiver operating curve (ROC) analysis while in other embodiments correlation analysis may be used. [0058] In block 1316, it is determined whether only the set of complex indicators C;, j= 1 to J, are to be considered in the model for suitability indicator, S. If so, then the method 30 proceeds to block 1320. Otherwise, the method proceeds to block 1318, where, one or more relationships are determined between each of the facial descriptor measurements, Fi, WO 2013/147932 PCT/US2012/064178 29 and the categorical additional data across all N subjects in the sample library. In addition, one or more relationships also are determined between each of the principal indicators, Pk, and the categorical additional data across all N subjects in the sample library. These relationships can be determined in the manner described above with respect to the 5 relationships with the principal and complex additional data. [0059] In block 1320, one or more of the facial descriptor measurements, Fi, principal indicators, Pk, and complex indicators, C 1 , are selected to be used to compute the suitability indicator, S. Any number of facial descriptor measurements, principal indicators, and complex indicators - up to and including the entire set of all three - may 10 be selected for each suitability indicator. Moreover, the selection may be limited only to one or more of the complex indicators, C;, i.e., the facial descriptor measurements and principal indicators are not used. The selection may be made in various ways, depending on the available information and the particular suitability indicator, S. The selection may be based on the raw values of Fi, Pk,, and/or Cj, or upon scaled or transformed versions. 15 For example, the z-score of a particular Fi, Pk,, or Cj, may be used for purposes of selection. Other transformations will be apparent to persons of ordinary skill in the art. [0060] In block 1320, the combination of the selected facial descriptor measurements, principal indicators, and complex indicators that optimizes suitability indicator S is also determined. In some embodiments, an optimal linear combination of the selected group 20 of Fi, Pk,, and/or C; is determined. In other words, if S = a1Fkl + cL2Fk2 + ... a.Fkn + i1Pki + p2Pk2 + ... pmPkm + 71Cki + 7 2 Ck2 + ... y rCkr then the facial descriptor measurements Fki, i = 1 to n, principal indicators Pki, i = 1 to m, and complex indicators, Cki i = 1 to r, are selected in block 1320 and the linear coefficients ai, i = 1 to n, Pi, i = 1 to m, and yi, i = 1 to r are determined in block 1320. To the extent that the selection is limited to one or more 25 of the complex indicators, C;, the corresponding coefficients ai and Pi are zero. The linear combination may be determined using statistical analysis techniques. For example, in in some embodiments, ROC analysis may be used to select the combination of facial descriptor measurements and/or principal indicators that provides a value for S that best categorizes the subjects in the sample library, e.g., according to at least a portion of the 30 categorical additional data. In other embodiments, correlation analysis may be used to select the combination of measurements and/or indicators. In the spirit of the disclosure, WO 2013/147932 PCT/US2012/064178 30 however, a non-linear combination of the selected group of facial descriptor measurements and/or principal indicators also may be determined based on the complex additional data, in the same or similar manner as discussed above. [0061] As mentioned above, the additional data stored in the sample library in block 5 1100 of Fig. 11 and block 1300 of Fig. 13 may comprise environmental data. An individual's actual personality and behavior is influenced not only by their innate personality traits but also by their environment, including experiences and social interactions throughout life, but particularly during early childhood. Accordingly, prediction of an individual's characteristic such as behavior or personality traits should 10 take into account the relevant environmental factors to the greatest extent possible. For example, the additional data stored in the sample library should include environmental information related to whether, or to what degree, each of the individuals has experienced particular events (e.g., death of a parent before a certain age), been exposed to certain environmental stimuli (e.g., violence), or has a particular family status (e.g., only child). 15 When determining a model for predicting a characteristic of an individual, such as by using one or more of the methods shown in Figs. 11 and 13, environmental variables may be incorporated in several different ways. First, environmental information about whether or not an individual has experienced a particular event can be represented by a Boolean-valued environmental variable. If this event is associated with one or more 20 facial descriptor measurements, principal indicators, or complex indicators, the environmental variable can be used to enable and disable the contributions of the associated measurements and/or indicators to the overall predictor model. For example, if environmental variable r is associated with principal indicator P 1 , then complex indicator C that depends on P1 would be expressed as C = T-a 1

-P

1 + a 2

P

2 + . anPn. If the 25 environmental information relates to degree or type of experience, the corresponding environmental variable (e.g., t) can take on either a discrete or a continuous range of values. [0062] In other embodiments, environmental variables may be included as standalone terms in the models determined for principal, complex, and/or categorical indicators. For 30 example, a complex indicator, C, that depends on environmental variable r as well as principal indicators P1 through Pn could be expressed as C = a 1

-P

1 + a 2

*P

2 + ... an.P, + WO 2013/147932 PCT/US2012/064178 31 an.1. ixThe value for the linear coefficient an,1 would then be determined together with linear coefficients ai through a,, e.g., in block 1314 of Fig. 13. More generally, in such embodiments, environmental variables may be treated in the same manner as facial descriptor measurements, principal indicators, and complex indicators when determining 5 the optimal selection and combination comprising a higher-level indicator. [0063] According to the embodiment illustrated by Fig. 12, a subset and combination of facial descriptor measurements selected using the embodiment illustrated by Fig. 11 then can be used to predict a characteristic of an individual based on the facial descriptor measurements for that individual. In other words, the subset and combination selected 10 based on the sample library can be applied to other individuals of the same type to determine the characteristic for those individuals. Although Fig. 12 illustrates the one or more embodiments by blocks arranged in a specific order, this order is merely exemplary and the steps or operations comprising the method may be performed in a different order than shown in the figure. Moreover, a person of ordinary skill will understand that the 15 blocks shown in Fig. 12 may be combined and/or divided into blocks having different functionality. [0064] In block 1200, digital images and additional data are obtained and stored for an individual of interest, in the same manner as described above with reference to the sample library (i.e., block 1100 of Fig. 11). In block 1202, reference points consistent with those 20 of the images in the sample library are added to the images of the individual of interest. In block 1204, facial descriptor measurements are calculated for the individual of interest. [0065] In block 1206, the subset and combination selected in block 1112 is applied to the facial descriptor measurements of the individual to predict the characteristic of interest. In block 1208, the sample library optionally may be augmented by adding the 25 image(s), additional data, and facial descriptor measurements for this individual. Subsequently, the method of Fig. 11 can be repeated using the augmented sample library, and the resulting predictor can be applied to additional individuals of the same type in accordance with the method of Fig. 12. Although the embodiment illustrated by Fig. 12 has been described as being used in conjunction with the embodiment of Fig. 11, it also 30 can be utilized in conjunction with the embodiment of Fig. 13. For example, the method WO 2013/147932 PCT/US2012/064178 32 shown in Fig. 12 can be used to compute one of more of the principal indicators, Pk that are comprised of a linear or non-linear combination of facial descriptor measurements, Fi. [0066] Fig. 14 is a flowchart of an alternative embodiment of a method for predicting a characteristic of an individual, such as suitability for a particular task, purpose, 5 relationship, etc., based on the facial descriptor measurements for that individual. The embodiment illustrated by Fig. 14 may utilize the facial descriptor measurements together with the models for principal indicators and complex indicators to determine a suitability indicator, S. The models for the principal and complex indicators may be determined by the method illustrated in Fig. 13 or by any other method that generates such models. 10 Although Fig. 14 illustrates the one or more embodiments by blocks arranged in a specific order, this order is merely exemplary and the steps or operations comprising the method may be performed in a different order than shown in the figure. Moreover, a person of ordinary skill will understand that the blocks shown in Fig. 14 may be combined and/or divided into blocks having different functionality. 15 [0067] In block 1400, digital images and additional data are obtained and stored for a subject, in the same manner as described above with reference to the sample library (e.g., block 1100 of Fig. 11 and block 1300 of Fig. 13). In block 1402, the images of the subject are annotated with a plurality of reference points. As discussed above with reference to Figs. 11 through 13, the annotation may be done manually or automatically. 20 In block 1404, a set of M facial descriptor measurements (Fi, i = 1 to M) are computed based on the reference points annotated onto the one or more images of the subject. [0068] In block 1406, it is determined whether principal indicators are to be used in the computation of the suitability indicator, S. In the case that only facial descriptor measurements Fi, i = 1 to M, are to be used in computing S, the method proceeds to block 25 1414. Otherwise, the method proceeds to block 1408, where principal indicators Pk, k = 1 to K, are computed using facial descriptor measurements Fi and the model determined, for example, in block 1308 of Fig. 13. Subsequently, the method proceeds to block 1410, where it is determined whether complex indicators are to be used to compute the suitability indicator. If not, then the method proceeds to block 1414. Otherwise, the 30 method proceeds to block 1412 where complex indicators Cj, j = 1 to J, are computed WO 2013/147932 PCT/US2012/064178 33 using facial descriptor measurements Fi and/or principal indicators Pk , based on the model determined, for example, in block 1314 of Fig. 13. [0069] Next, in block 1414, the suitability indicator, S, for the subject of interest is computed based on the facial descriptor measurements Fi, principal indicators Pk, and/or 5 complex indicators Cj using the model determined, for example, in block 1320 of Fig. 13. In block 1416, the subject is categorized based on the value of suitability indicator, S, and a suitability threshold, Sx. In some embodiments, this characterization may comprise determining that the subject is suitable for a particular task, purpose, relationship, etc. based on comparing the values of S and S, e.g., S > Sx or |S - S,| < E. In some 10 embodiments, this characterization may comprise assigning the subject to one category if S > Sx and another category if S < S,. Other possible comparisons will be apparent to persons of ordinary skill in the art. Finally, in block 1418, the sample library optionally may be augmented by adding the image(s), additional data, facial descriptor measurements, principal indicators, complex indicators, and/or suitability indicator for the 15 subject. [0070] Fig. 15 further illustrates the method of Fig. 14 by example. Initially, principal indicators P1, P 2 , and P 3 are computed based on linear combinations of facial descriptor measurements F 1 , F 2 , and F 3 . The use of the same facial descriptor measurements in computing each of the principal indicators is merely exemplary. Coefficients ai, Pi, and yi 20 used to compute the principal indicators may have been previously determined, for example, by the method shown in Fig. 13. Z-scores Z 1 , Z 2 , and Z 3 , are determined for principal indicators P 1 , P 2 , and P 3 , respectively. Next, complex indicators C 4 and C 5 are computed by a non-linear combination of principal indicator z-scores Z 1 , Z 2 , and Z 3 . Coefficients ai, Pi, yi, and pi used to compute the complex indicators may have been 25 previously determined, for example, by the method shown in Fig. 13. Z-scores Z 4 and Z 5 are determined for complex indicators C 4 and C 5 , respectively. [0071] Subsequently, categorical indicator for suitability for a task Stask is computed from a non-linear combination of z-scores Z 1 , Z 2 , and Z 3 corresponding to the principal indicators and z-scores Z 4 and Zs corresponding to the complex indicators. Coefficients ai, 30 pi, yi, and pi used to compute the suitability indicator Stask may have been previously determined, for example, by the method shown in Fig. 13. Next, Stask may be used to WO 2013/147932 PCT/US2012/064178 34 determine whether the subject with facial descriptor measurements F 1 , F 2 , and F 3 is suitable for the task at issue. This may be done in a variety of ways, two of which are shown in Fig. 15. For example, the raw value of Stask may be compared against the threshold S, to determine which of two categories to place the subject (e.g., "(+)" and "( 5 )" in Fig. 15). In another example, a z-score Ztask computed from Stak may be compared to a cutoff or threshold on the distribution of the population of interest (e.g., the 80th percentile) to determine whether the subject is suitable for the task of interest. The population of interest may be the general population or a specific sub-population, e.g., those who already are engaged in the particular task. The cutoff or threshold may be 10 chosen in conjunction with the choice of population. As noted above, Fig. 15 is merely an example illustrating the embodiment shown in and described above in reference to Fig. 14; other variations will be apparent to the person of ordinary skill. [0072] In other embodiments, Stask may comprise a set of z-scores, Zq, q=1 to Q, each corresponding to one or more principal indicators Pk and/or one or more complex 15 indicators Cj, and threshold Sx may comprise a set of thresholds, Sxq, q=1 to Q, each corresponding to a Zq. In such embodiments, each of z-scores in (Zq} can be compared to its corresponding threshold in {Sxq}, with subject categorized as suitable for the task of interest if each member of {Zq} has the desired relationship - less or greater than, as the case may be - with the corresponding member of {Sxq}. Alternately, the subject may be 20 determined to be suitable if a subset of (Zq} and {Sxq} have the desired relationship (e.g., majority). In some embodiments, one or more of the thresholds {Sxq} may comprise a pair of thresholds related to the corresponding member of {Zg}, in which case the desired relationship may be between the pair of thresholds, e.g., Sxqi < Zq < Sxq2. In some embodiments, Stask may comprise a set of raw values, (Sq}, each corresponding to one or 25 more principal indicators Pk, and/or one or more complex indicators Cj, and threshold S, may comprise a set of thresholds {Sxq}, each corresponding to a member of (Sq}. Suitability may be determined for these embodiments in the same manner as discussed above with respect to embodiments comprising z-scores. [0073] In the embodiments described above, the operation in block 1416 of Fig. 14 30 categorizes the subject as suitable or non-suitable based on the suitability indicator. In other embodiments, however, the operation in block 1416 may comprise using the WO 2013/147932 PCT/US2012/064178 35 suitability indicator to assign the subject to one of a plurality, or set, of categories. This set of categories may be exhaustive (i.e., every member of a population will be placed in one of the categories) or non-exhaustive, in which case the set may comprise an "other" or similarly-named category to which subjects that do not fall into one of the defined 5 categories are assigned. In some embodiments, a set of categories Kt, t = 1 to T, are defined by a set of thresholds, Sxt, t = 1 to T+1, such that a subject is assigned to category Kt if Sxt < Stask Sx(t+l). Optionally, beginning threshold Sx 1 and end threshold Sx(T+1) may be set to -/+oo, respectively. [0074] In other embodiments, Stask may comprise a set of z-scores, Zq, q=1 to Q - or 10 raw scores, Sq, q=1 to Q - each corresponding to one or more principal indicators Pk and/or one or more complex indicators C;. Each category Kt is defined by a pair of values {Sxqt"i, Sxqt"x} defining a range for each z-score of the set (Zq}, i.e., a set of Q ranges. A subject will be assigned to a category Kt if each member of the set of z-scores {Zq} falls within the range { S Sxqtmax } associated with that z-score and category. In other 15 words, a subject will be assigned to a particular category only if all z-scores (or raw scores, as the case may be) fall within the set of ranges associated with that category. In other embodiments, a subject may be assigned to a particular category if a required subset of the z-scores (or raw scores) falls within the set of ranges associated with that category. The required subset may be determined in various ways, such as by majority (i.e., at least 20 Q/2) of the set, by majority of each of two subsets, by all of a one subset and one or more of the remainder, etc. In these embodiments, the set of categories { Kt } may be exhaustive (i.e., every individual in the population will be assigned to one of the categories) or non exhaustive. [0075] For example, embodiments of the present disclosure may be used to predict the 25 risk level of a driver for purposes of automobile insurance. Accordingly, Stask may comprise a set of z-scores, Zq, q=1 to Q - or raw scores, Sq, q=1 to Q - each related to a particular risk area for a driver (e.g., specific types of accidents, theft risk, etc.). In some embodiments, the value of a particular z-score Zq is proportional to the risk level in the corresponding risk area. An insured driver (or candidate for insurance) could be assigned 30 to one of a plurality of overall risk categories { Kt, t = 1 to T } based on the values of each of the set of z-scores {Zg} relative to one or more corresponding thresholds. In such WO 2013/147932 PCT/US2012/064178 36 embodiments, additional data used to determine the model for categorization of drivers may comprise frequency and/or severity of past accidents, overall cost of accidents to insurance company, number of traffic violations, traffic density in individual's home city, length of commute (if any), etc. In some embodiments, such a model may be applied 5 using additional data and facial descriptor measurements pertaining to a new individual to predict that individual's risk level. [0076] In other embodiments, the set of categories may be related to a pre-existing population categorization model, such as a pre-existing model used by insurance companies to categorize drivers and personality categorization models such as the Myers 10 Briggs Type Indicator (MBTI), the Kiersey Temperament Sorter, the Costa McRae Five Factor model (also known as the Big Five model), the Petrides Trait El model for emotional intelligence, and others known to persons of ordinary skill in the art. For example, the MBTI separates individuals into sixteen personality categories using subjects' answers to a psychometric questionnaire designed to measure psychological 15 preferences in how people perceive the world and make decisions. The MBTI categories are based on the four dichotomies of extraversion/introversion (E/I), sensing/intuition (S/N), thinking/feeling (T/F), and judging/perceiving (J/P). [0077] The method shown in Fig. 13 could be employed to predict which of the 16 categories that an individual would be assigned by an MBTI assessment. For example, 20 the sample library created in block 1300 may comprise digital images of individuals whose personalities have been categorized using MBTI, together with additional data for each individual comprising their answers to the questionnaire, their rating on each of the four dichotomies, and any other information used for MBTI categorization (e.g., gender). In blocks 1302 through 1318, a variety of facial descriptor measurements, principal 25 indicators, and complex indicators could be determined as described above. In some embodiments, a set of four predictors {SE,,, SS/N, ST/F, SJ/P) corresponding to the four MBTI dichotomies are determined in block 1320. The set of predictors may be based on raw scores, z-scores, or a combination thereof. An indicator of an individual's MBTI category can be determined based on this set of dichotomy predictors. 30 [0078] These predictors may then be used according to the method shown in Fig. 14 to predict to which of the sixteen personality types that an individual subject would be WO 2013/147932 PCT/US2012/064178 37 assigned based on an MBTI assessment. The subject's predicted rating on the four MBTI dichotomies can be determined in block 1416 by comparing the subject's four dichotomy indicators {SE/I, Ss/N, ST/F, Sj/p) to their corresponding thresholds {S,,E 1 , Sxs/N, SxT/F, SxJ/P} By way of example, if SEi < SxEr, then the subject is likely to be categorized as an 5 introvert ("I"); otherwise the subject is likely to be categorized as an extrovert ("E") by an MBTI assessment. The subject's ultimate MBTI category is predicted based on the results of the four comparisons. In another embodiment, a single category predictor SMBTI may be determined in block 1320 and used to compute a single categorical indicator for a subject in block 1416. An indicator of the subject's MBTI category may be determined 10 based upon the range of values that the categorical indicator falls within, as described above. [0079] While some embodiments of the present disclosure are useful for predicting categorizations of individuals under existing models, such as the MBTI, other embodiments can be used to augment existing categorization models with additional, 15 more granular information. For example, in addition to predicting a subject's rating on each of the four MBTI dichotomies, the methods of Figs. 11 through 15 can be used to determine indicators of the strength of the subject's rating on each of the categories. The four dichotomy indicators {SE/, Ss/N, ST/F, SJ/P} may be expressed as z-scores such that a high positive value for a dichotomy (e.g., E/I) indicates that the subject is strongly 20 oriented toward one side (e.g., E), a high negative value indicates that the subject is strongly oriented toward the other side (e.g., I), and a value near zero indicates that the subject is relatively balanced. Alternately, the four dichotomy indicators may be expressed as percentile scores, e.g., with zero and 100 percentiles representing the two extremes of a dichotomy (e.g., extreme introverts and extreme extroverts). Continuously 25 valued indicators, such as these examples, may be used to create sub-categories for one or more of the categories of the MBTI or other existent model. [0080] Although exemplary embodiments have been discussed above in relation to humans, embodiments of the methods for predicting a characteristic of an individual may be applied for predicting various characteristics of a non-human animal. For example, 30 embodiments may be employed to predict the characteristic of gluttony in animals, i.e., how much a specific animal will eat in relation to the population of other animals of the WO 2013/147932 PCT/US2012/064178 38 same type. Moreover, such embodiments may be employed to predict how the animal's food intake level is affected by environmental stress (e.g., heat, handling, transport, weaning, parturition, injury, etc.) based on the characteristics of their susceptibility to stress and their ability to respond to it. Furthermore, embodiments may be employed to 5 predict an animal's preferences for particular types of food (based, e.g., on texture, taste, smell, etc.) and how this contributes to variability in the animal's food intake level. Embodiments also may be used to predict how well an animal is able to digest their food. [0081] Similarly, embodiments may be employed to predict how an animal will respond to changes in their environment such as, for example, changes to pen or stall, pen- or 10 stallmates, transportation, feeding systems, feeding or other schedules, etc. Embodiments may be used to predict the animal's susceptibility to stress, the expected level of stress the animal would suffer in response to various stress-inducing changes, and type and/or degree of physiological reaction to such changes including depression, aggression, flight, weight loss, diarrhea, self-injury, death, etc. Based on the predictions of such 15 characteristics, such embodiments could be used to select animals better suited to thrive in terms of growth and/or production under the known stress of existing animal management protocols. [0082] Although the methods of Figs. 11 through 15 have been described above in terms of determining a predictor of a characteristic or indicator of an individual (or 20 predicting the characteristic or indicator using the predictor, as the case may be), embodiments of the present disclosure also may be used to determine the degree to which two individuals are suitable for a particular purpose or type of relationship, i.e., "pairwise complementary." These embodiments may be used to determine if two individuals are pairwise complementary for any type of relationship, including husband-wife, student 25 teacher, athlete-coach/trainer, teammates, roommates, worker-supervisor, co-workers, business partners, etc. [0083] For example, Fig. 16 shows a flowchart for a method for determining a predictor for the pairwise complementarity of two individuals for a particular purpose, e.g., marriage. As will be apparent during the following discussion, the embodiment 30 illustrated by Fig. 16 is similar to the embodiment shown in and described above with reference to Fig. 13. Although Fig. 16 illustrates the one or more embodiments by blocks WO 2013/147932 PCT/US2012/064178 39 arranged in a specific order, this order is merely exemplary and the steps or operations comprising the method may be performed in a different order than shown in the figure. Moreover, a person of ordinary skill will understand that the blocks shown in Fig. 16 may be combined and/or divided into blocks having different functionality. 5 [0084] In block 1600, a sample library comprised of data for N individuals or subjects is created. These N individuals are comprised of N/2 pairs of individuals that are associated based on an existing relationship of the type for which a predictor will be determined. For instance, the N individuals may comprise N/2 married couples. The operation comprises obtaining one or more digital images of an anatomical region for 10 each of the individuals, as discussed above with reference to block 1100 of Fig. 11 and block 1300 of Fig. 13 with block 1600 also storing additional data related to the individuals and their pairings in the sample library. This additional data may comprise individual additional data, such as discussed above with reference to block 1100 of Fig. 11 and block 1300 of Fig. 13, and pairwise additional data, which relates to the pairing of 15 the two individuals. The pairwise additional data may comprise identifiers for the two individuals that are paired, quantitative information related to the duration, success, performance, etc. of the pairing, and environmental data related to the pairing of the individuals (e.g., existence and/or degree of certain external events or stimuli). The pairwise additional data may relate specifically to individual pairs in the sample library, a 20 plurality of pairs in the sample library, or more generally to a population of which the pairs are a part. The additional data is stored in a manner and location by which it can be associated with the sample library images, such as in a relational database accessible by a digital computer. [0085] In block 1602, the images corresponding to each of the N individuals or subjects 25 in the sample library are annotated with a plurality of reference points. Also in block 1602, a set of M facial descriptor measurements (Fi, i = 1 to M) are computed based on the reference points annotated onto the one or more images for each of the N individuals. In block 1604, one or more of the facial descriptor measurements, Fi, are selected to be used to compute each of individual principal indicators Pk, k = 1 to K. Also in block 30 1604, the combination of the selected facial descriptor measurements that optimizes each of individual principal indicators Pk is also determined. In block 1606, one or more of the WO 2013/147932 PCT/US2012/064178 40 facial descriptor measurements, Fi, and individual principal indicators, Pk, are selected to be used to compute each of individual complex indicators C 1 , j = 1 to J. Also in block 1606, the combination of the selected facial descriptor measurements, Fi, and/or individual principal indicators Pi that optimizes each of individual complex indicators C 5 is also determined. The operations in blocks 1602, 1604, and 1606 may be substantially the same as the operations described above with reference to blocks 1302 through 1314 of Fig. 13. Furthermore, the operation in block 1606 may be omitted if no complex indicators are utilized in the prediction model being determined. In such embodiments, the method proceeds from block 1604 to block 1608. 10 [0086] In block 1608, for each of the principal indicators Pk, one or more relationships are determined between the pairwise additional data and one or more functions f(PkX, Pe) of the individual principal indicator Pk for an associated pair of individuals (x, y), across all N/2 associated pairs in the sample library. Also in block 1608, for each of the complex indicators Cj, one or more relationships are determined between the pairwise 15 additional data and one or more functions f(Cf, CY) of the individual complex indicator C for the associated pair (x, y), across all N/2 associated pairs in the sample library. The functions f(P, Pe) and f(CJ, C/) may be any linear or non-linear function of the pair of the particular individual indicator for both individuals in the pair. Exemplary functions f (Pk', Pe') include, but are not limited to, PkX, Pe, (a.Pk' + P Pe), (a-PkX - 0-Pe), 1/tPkX - PI, 20 max(Pkj, Pe), (Px *Pe), (Pkxl/PkY), etc., and like functions for f (Cf, Cf). Some embodiments also may include discontinuous functions. In some embodiments, the operations of block 1608 may comprise determining one or more relationships between functions of multiple individual principal or complex indicators for pairs of individuals, e.g.,f (Pk1X, P1Y, Pk2, P') for individual principal indicators Pki and Pi2. 25 [0087] In block 1608, for each of the principal and complex indicators, the set of pairwise functions f(Pkx, Pe) and f(Cf, C/) to be analyzed for a relationship with the pairwise additional data may be determined in various ways. In some embodiments, a fixed set is always analyzed, e.g., PkX, Pe, (a-Pkx + P-Pk), (a-Pkx - .Pk), 1/IPkX - Pke, max(Pkx, Pe), (Px Pe), and (PxlPe), for relationships with the additional data. One or 30 more of the functions in the fixed set may be selected for each principal and/or complex indicator based on categorical potential with respect to the additional data using, e.g., WO 2013/147932 PCT/US2012/064178 41 ROC or correlation analysis as described above. In other embodiments, one or more candidate functions of each principal and/or complex indicator may be selected by using heuristic computational methods. For example, a larger list of candidate functions may be organized into a multi-stage search tree. At each stage, some small number of (e.g., two) 5 candidate functions - or types of functions - are analyzed to determine which gives the best categorical potential based on the additional data. The best of the candidates is selected and used in the next stage of the search tree. In this manner, a small set of candidate functions of each of the individual principal and/or complex indicators can be identified from among a larger set of candidate functions. Such heuristic computational 10 algorithms also may comprise artificial intelligence functions in which results of previous selections related to the same characteristic or indicator may be stored and used to guide future selections, as known by persons of ordinary skill in the art. [0088] The operation in block 1608 of selecting the pairwise functions f(Pkx, Pk') and f(Cf, C; Y) may be broken into steps that are carried out sequentially. In some 15 embodiments, the operation of block 1608 may comprise first selecting one or more pairs of principal indicators (Pk', P2') and/or complex indicators (Cf, Cf) then determining the optimal functions f(Pk', P2j) and f(Cf, C/) of the selected pairs. In other embodiments, the selection of the indicator pairs and the determination of the optimal pairwise functions may comprise a substantially unitary operation. Other ways of structuring the operations 20 of block 1608 will be apparent to persons of ordinary skill in the art. Regardless of the selection process used, however, the candidate pairwise functions analyzed may comprise functions of individual principal indicators Pk' and/or Pk', and functions of individual complex indicators Cf and/or Cf. As used herein unless specifically stated to the contrary, the term "pairwise function" comprises functions of individual principal or 25 complex indicators, or other such terms. Similarly, a function of two variables f(x, y) comprises functionsf(x) andfty) of the individual variables. [0089] Moreover, although the operation of block 1608 is described above in the context of finding pairwise functions of the same type of indicator (i.e., principal or complex), this is merely exemplary and pairwise functions of different types of indicators 30 may be selected in some embodiments. For example, embodiments of block 1608 may select one or more functions f(Pk(, Pe, Cj, Cf) of both principal and complex indicators WO 2013/147932 PCT/US2012/064178 42 for the two individuals x and y. Furthermore, embodiments of block 1608 may select one or more functions of a combination of principal indicators, complex indicators, and facial descriptor measurements, Fi, corresponding to the pair of individuals x and y. Accordingly, although the description of operations of subsequent blocks of Fig. 16 may 5 refer to pairwise functions f(Px, P') and/or f(C;, CY), persons of ordinary skill in the art will understand that these operations may encompass functions of various indicators and measurements pertaining to two individuals x and y. [0090] In any event, the result of block 1608 is one or more pairwise functions corresponding to the individual principal and/or complex indicators, together with their 10 respective relationships to the pairwise additional data. In block 1610, some portion of the group of the pairwise functions determined in block 1608 are selected to be used to compute the pairwise complementarity indicator, S. Any number of the pairwise functions - up to and including the entire set - may be selected. Moreover, the selection may be limited only to functions f(Cj x, Cj Y) of individual complex indicators, i.e., 15 functions f(Pkx, Pej) of individual principal indicators are not used. The selection may be made in various ways, depending on the available information and the particular pairwise complementarity indicator, S. The selection may be based on the raw values of functions f(Pk', Pe) and/or f(Cj X, CjY), or upon scaled or transformed versions. For example, the z score of a particular functions f(Pkx, Pke) or f(Cf, C/) may be used for purposes of 20 selection. Other transformations will be apparent to persons of ordinary skill. [0091] The selection of the pairwise functions f(Pkx, Pk7) and/or f(Cj, C/) to be used for computing indicator S may be accomplished in various ways. In some embodiments, some fixed number of pairwise functions having the highest categorical potential with respect to the additional data may be selected. In other embodiments, one or more 25 pairwise functions may be selected by using heuristic computational methods. For example, the various functions can be broken into groups or categories using qualitative information related to the types of characteristics that they indicate. Subsequently, one or more pairwise functions may be selected from each group based on, e.g., their categorical potential within the group. Alternately, selections may be constrained to a subset of the 30 groups based on the qualitative information related to the groups. Such embodiments may comprise learning-based functionality whereby results of previous selections related WO 2013/147932 PCT/US2012/064178 43 to the same or similar indicator(s) may be stored and used to guide future selections. Other methods for selecting pairwise functions f(Pkx, Pe) and/or f(Cf, C/) to be used for computing indicator S will be apparent to the person of ordinary skill in the art. [0092] In block 1610, a combination of the selected pairwise functions f(Pkx, PI') and 5 f(Cf, Cf) that optimizes pairwise complementarity indicator S is also determined. In some embodiments, an optimal linear combination of the selected pairwise functions f(Pkx, Pkj) and f(C x, C Y) is determined. In other words, if S = i f i (pi, Ply) + a 2 f 2

(P

2 X,

P

2 Y) + ... a , f,(Pnx, Pny) + 0 1 fi(Clx, C 1 l) + 02 f 2

(C

2 x, C?) + ... m fm(Cmx, CmY) then the pairwise functions f(PI(, Ply), i = 1 to n, and f(C, x, Ci Y), i = 1 to m, are selected and the 10 linear coefficients ai, i = 1 to n, and j, i = 1 to m, are determined in block 1610. To the extent that the selection is limited to one or more of pairwise functions f(CIt, Ci Y), the corresponding coefficients ai are zero. The linear combination may be determined using statistical analysis techniques. For example, in categorical models, ROC analysis may be used to select the combination of pairwise functions f(Px, PI) and/or f(Cf, C/) that 15 provides a value for S that best predicts or categorizes the pairs in the sample library, e.g., according to at least a portion of the pairwise additional data. In the spirit of the disclosure, however, a non-linear combination of the selected group of pairwise functions f(PkX, Pkj) and/or f(Cj, C Y) also may be determined based on the pairwise additional data. A non-linear combination of the selected group of pairwise functions f(Pkx, Pj) and/or 20 f(Cf, Cj) may be determined using optimization techniques including, but not limited to, Newton's Method and Lagrange's Method. The selected linear and/or non-linear combination may be of the pairwise functions f(Pkx, Pk') and/or f(Cf, C/) themselves, or of scaled or transformed versions such as z-scores. [0093] Although not shown in Fig. 16, persons of ordinary skill will readily 25 comprehend understand that the model may comprise one or more additional levels of indicators. For example, the model may comprise one or more "tertiary indicators" that may be based on, or combinations of, facial descriptor measurements, principal indicators, and/or complex indicators. In such case, the one or more suitability indicators may be based on any of the facial descriptor measurements, principal indicators, complex 30 indicators, and tertiary indicators. Moreover, models comprising such additional levels of indicators may be used in other methods described herein for predicting characteristics of WO 2013/147932 PCT/US2012/064178 44 one or more individuals, such as shown in and described below with reference to Figs. 17 and 18. [0094] Fig. 17 is a flowchart of an embodiment of a method for predicting a characteristic of a pair of individuals, such as pairwise complementarity for a particular 5 purpose (e.g., marriage), based on the facial descriptor measurements for those individuals. The embodiment illustrated by Fig. 17 may utilize models for principal indicators, complex indicators, and pairwise indicator functions to determine a pairwise complementarity indicator, S. The models for principal indicators, complex indicators, and pairwise indicator functions may be determined by the method illustrated in Fig. 16 10 or by any other method that generates such models. Although Fig. 17 illustrates the one or more embodiments by blocks arranged in a specific order, this order is merely exemplary and the steps or operations comprising the method may be performed in a different order than shown in the figure. Moreover, a person of ordinary skill will understand that the blocks shown in Fig. 17 may be combined and/or divided into blocks having different 15 functionality. [0095] In block 1720, digital images and additional data are obtained and stored for a pair of individuals, in the same manner as described above with reference to the sample library (e.g., block 1600 of Fig. 16). In block 1722, the images of the subject are annotated with a plurality of reference points. In block 1724, a set of facial descriptor 20 measurements, Fi, i = 1 to M, are computed based on the reference points annotated onto the one or more images of the subject. In block 1726, individual principal indicators Pk, k = 1 to K, are computed using facial descriptor measurements F, and the individual principal indicator model determined, for example, in block 1704 of Fig. 16. In block 1728, individual complex indicators Cj, j = 1 to J, are computed using facial descriptor 25 measurements F, and/or individual principal indicators Pk based on the model determined, 1 2 for example, in block 1706 of Fig. 16. In block 1730, pairwise functions f(Pk , Pk ) and/or f(CJ', C.

2 ) of principal and/or complex indicators for the two individuals (i.e., individuals 1 and 2) are computed, and further used to compute pairwise complementarity indicator, S, of the prospective pair using the model determined, for example, in block 1610 of Fig. 30 16.

WO 2013/147932 PCT/US2012/064178 45 [0096] Next, in block 1732, the subject is categorized based on the values of the pairwise complementarity indicator, S and a suitability threshold, Sx. In some embodiments, this characterization may comprise determining that the pair of individuals are suitable for a particular purpose or relationship (e.g., marriage) based on comparing 5 the values of S and Sx, e.g., S > Sx or |S - Sx < c. In some embodiments, this categorization may comprise assigning the pair to one category if S > Sx and another category if S < Sx. Other possible comparisons will be apparent to persons of ordinary skill in the art. Finally, in block 1734, the sample library optionally may be augmented by adding the image(s), additional data, facial descriptor measurements, principal 10 indicators, complex indicators, pairwise functions, and/or pairwise complementarity indicator for the pair of individuals. [0097] In other embodiments, the operation in block 1732 may comprise using the pairwise complementarity indicator to assign the pair to one of a plurality, or set, of categories. In some embodiments, a set of categories Kt, t = 1 to T, are defined by a set of 15 thresholds, Sxt, t = 1 to T+1, such that a subject is assigned to category Kt if Sxt < S < Sx(t+1). Optionally, beginning threshold Sx, and end threshold Sx(T11) may be set to -/+co, respectively. In other embodiments, S may comprise a set of z-scores, Zq, q=1 to Q - or 1 2 raw scores, Sq, q=1 to Q - each corresponding to pairwise functions f(Pk , Pk) and/or f(Cj 1 , CJ) of principal and/or complex indicators for the two individuals. In such 20 embodiments, each of the set of categories Kt, t = 1 to T, may be defined by a particular range {Sxqti, Sxqt2} associated with a particular z-score, Zq. A pair of individuals will be assigned to a category Kt if each member of the set of z-scores {Zq} falls within the particular range {Sxqti, Sxqa} associated with that z-score and category. In other embodiments, a pair of individuals may be assigned to a particular category if a required 25 subset of the z-scores (or raw scores) falls within the set of ranges associated with that category. For example, an on-line dating service could categorize the type of relationship that two individuals are most likely to experience. If the type of relationship that each of the individuals is seeking was included in the individual additional data, such embodiments could also use it to determine the likelihood of success of the prospective 30 pairing.

WO 2013/147932 PCT/US2012/064178 46 [0098] Variations of the embodiment shown in and described above with reference to Fig. 17 are also possible, as will be apparent to persons of ordinary skill in the art. For example, rather than predicting whether two new individuals (i.e., individuals not in the sample library) are pairwise complementary, the method of Fig. 17 may also be used to 5 predict which of the existing members of a sample library would be pairwise complementary with a single new individual. Embodiments also may be used to determine the degree to which a new individual would be pairwise complementary with one or more existing members of the sample library, or to identify or rank a predetermined number of existing members of the sample library based on their degree of 10 pairwise complementarity with the new individual. [0099] In such embodiments involving pairings of a single new individual with one or more existing individuals in the sample library, the operations in blocks 1720 through 1728 and 1736 of Fig. 17 will be performed only for the new individual, provided that the data records of existing individuals in the sample library are complete. Furthermore, the 15 operations in blocks 1730 and 1732 will be performed for each pairing of the new individual with a different existing individual in the sample library. The operations of block 1734 may comprise the ranking or ordering of the existing individuals tested in blocks 1730 and 1732 based on their degree of pairwise complementarity with the new individual. 20 [00100] Other embodiments of Figs. 16 and 17 may be used to categorize an existing relationship of two individuals (e.g., a marriage) in various ways. For example, the pairwise complementarity indicator, S, may comprise a set of z-scores {Zq}, each corresponding to one or more pairwise functions of principal and/or complex indicators for the two individuals. The set of one or more pairwise functions may be selected, as 25 shown in Fig. 16, such that the corresponding Zq represents or is related to a specific area in the particular type of relationship of interest. Embodiments may categorize each specific area of the relationship by comparing the value of its corresponding z-score Zq to one or more thresholds Sxq, and assigning a quantitative or qualitative category accordingly. For example, embodiments may compare Zq to a series of thresholds { Sxqi, 30 Sxq2, ... Sxqn}, each representing a boundary between two of the n+1 equal-probability sections of the probability distribution of the one or more pairwise functions WO 2013/147932 PCT/US2012/064178 47 corresponding to Zq (e.g., quartiles for n = 3). Subsequently, the area of the relationship corresponding to Zq will be categorized based on which of the equal-probability sections that Zq falls within. This categorization may be quantitative (e.g., first quartile, sixth decile) or qualitative (e.g., very weak, weak, average, strong, very strong), or a 5 combination thereof. (e.g., low second quartile). [00101] In addition, various embodiments of the present disclosure may be used to determine whether, or the degree to which, a group of more than two individuals are suitable for one or more purposes, tasks, and/or relationships. For example, embodiments may be used to determine if a group of individuals is suitable for a specific purpose, task, 10 or relationship. A group suitability indicator, SG, corresponding to the specific task may be determined based on linear and/or non-linear combinations of the individual principal and/or complex indicators, or may be a statistic of the individual indicators. For example, if the set {C 1 ', i = 1 to I} comprises complex indicator C1 for each of the I individuals in the group, the group categorical indicator may be based on mean, median, mode, standard 15 deviation, variance, ratio of standard deviation to range, average distance between numerically-ordered values, etc. of the values in {C1'}. The group suitability indicator SG may comprise a z-score, raw score, percentile score, or other metric indicating whether, or to what extent, the group is suitable for the specific task. For example, a percentile score may indicate the suitability ranking of a particular group among all groups engaged 20 in the same task. The suitability of the group for the task may be determined, for example, based on the relationship of SG to a threshold value Sx, as described above with reference to other embodiments. [00102] Similarly, embodiments may be used to determine which one of a particular set of tasks is most suitable for a particular group of individuals. A group categorical 25 indicator, SG, may be determined using principal and/or complex indicators for individuals in the group. The group categorical indicator may be based on linear and/or non-linear combinations of the principal and/or complex indicators, or may be a statistic of the individual indicators. The group categorical indicator may further comprise a set of categories {Kt, t = 1 to T}, that are defined by a set of thresholds (Sxt, t = 1 to T+1}, 30 such that the group is assigned to category Kt if Sxt < SG Sx(t+1). Optionally, beginning threshold Sxi and end threshold Sx(T+1) may be set to -/+oo, respectively. Each of the WO 2013/147932 PCT/US2012/064178 48 categories Kt may correspond to a particular task, and the most suitable task for the group may be determined based on the category Kt to which the group is assigned. [00103] In other embodiments, a set of group suitability indicators {SGt, t = 1 to TI, each corresponding to one of the particular set of tasks, may be determined based on 5 linear and/or non-linear combinations of the individual principal and/or complex indicators, or may be a statistic of the individual indicators. The set of group indicators {SGt}may comprise a set of z-scores, raw scores, percentile scores, or other metrics that indicates whether, or to what extent, the group is suitable for each of the corresponding set of tasks. For example, a percentile score for a task t may indicate the suitability 10 ranking of a particular group among all groups engaged in task t. The most suitable task for the group may be determined, for example, based on selecting the task corresponding to the highest group suitability indicator among the set {SGt}. [00104] Other embodiments of the present disclosure may be used to determine whether, or to what degree, an existing or prospective group of individuals fills a 15 predetermined set of roles. The set of roles may be defined based on what are deemed necessary for a particular task or purpose, or may be a more generic, or broader, set of roles not oriented or directed toward any particular task or purpose, e.g., "leader", "facilitator", "innovator", "executor", etc. Suitability indicator models can be defined for each role, based on principal and/or complex individual indicators related to 20 characteristics associated with the respective roles (e.g., sensitivity, dominance, impulsiveness, creativity, etc.). Using these models and the facial descriptor measurements of the individuals comprising the group, a set of suitability indicators {S/, r = 1 to R} corresponding to the set of R roles may be computed for each individual i in the group. The suitability of an individual for a specific role may be determined by 25 comparing {Sr t } to a corresponding threshold to determine a "fit/no fit" rating. Alternately, or in addition, the suitability indicator for the individual/role combination may comprise degree-of-fit information, which may be categorical, ordinal, integer valued, real-valued, etc. If an individual's indicators {Sr 1 } show that they are suitable for multiple roles, they may be assigned to each suitable role category. If individual's 30 suitability indicators show that they are suitable for no roles, they may be assigned to a "none" role category or to the role category having the greatest degree-of-fit, provided WO 2013/147932 PCT/US2012/064178 49 such information is available. Alternately, a single categorical indicator model may be determined and used to assign the individual to one category based on the value of their corresponding indicator, in the same manner as described above with respect to group categorical indicators. 5 [00105] The group suitability indicator may be determined in various ways. It may be based on some function of the number of individuals in each role category, number or percentage of role categories unoccupied, number or percentage of individuals in the "none" role category, etc. Alternately, if individual suitability indicators comprise degree-of-fit information, they may be used to determine the group suitability indicator. 10 For example, if a particular distribution among role categories is preferred (e.g., for a particular task), the group of individuals can be mapped to this preferred distribution in a way that maximizes the group suitability indicator, which may be computed by a linear (e.g., sum or weighted sum) or non-linear combination of the individual suitability indicators corresponding to the individual/role category mapping. Such mapping could 15 be constrained, for example, by requiring all, a portion, or a specific subset of the role categories to be filled by at least one individual. [00106] Other embodiments of the present disclosure may be used to determine whether, or to what degree, a new individual fits within an existing group of individuals for a specific task, purpose, or relationship. In some embodiments, suitability of the new 20 individual as a member of the group may be determined by constraining the roles of the existing group members to remain the same, and then using facial descriptor measurement analysis techniques to predict whether the new individual will fill a defined missing role within the group. In these embodiments, a set of suitability indicators {Sr, r = 1 to R} corresponding to the set of R missing roles may be computed for the 25 prospective new group member, in the same manner as described above. The suitability of an individual for a specific role may be determined by comparing {S,} to a corresponding threshold to determine a "fit/no fit" rating. Alternately, or in addition, the set of suitability indicators may comprise degree-of-fit information, which may be categorical, ordinal, integer-valued, real-valued, etc. If the new individual's indicators 30 {S,} show that they are suitable for one or more of the missing roles, they may be assigned to each suitable role category.

WO 2013/147932 PCT/US2012/064178 50 [00107] In other embodiments, the new individual's suitability may be based on the existence, or degree, of conflict between the predicted role(s) for the new individual and roles and/or categories of existing group members. The existence, or degree, of conflict may be determined, for example, based on a pairwise conflict indicator between the new 5 individual and each of the existing group members. Such an indicator may be computed in a manner similar to the pairwise complementarity indicators discussed above. [00108] Fig. 18 shows an exemplary method for determining the new individual's suitability be based on the existence, or degree, of conflict between the predicted role(s) for the new individual and roles and/or categories of existing group members. Although 10 Fig. 18 illustrates the one or more embodiments by blocks arranged in a specific order, this order is merely exemplary and the steps or operations comprising the method may be performed in a different order than shown in the figure. Moreover, a person of ordinary skill will understand that the blocks shown in Fig. 18 may be combined and/or divided into blocks having different functionality. 15 [00109] In block 1800, digital images and additional data are obtained and stored for the new individual, i.e., the prospective group member. The images of the new individual are annotated with a plurality of reference points and a set of facial descriptor measurements, Fi, i = 1 to M, are computed based on the annotated reference points. In addition, principal indicators Pk, k = 1 to K, are computed using facial descriptor 20 measurements Fi, and individual complex indicators Cj, j = 1 to J, are computed using facial descriptor measurements Fi and/or individual principal indicators Pk. The models used to compute the principal and complex indicators may be determined, for example, as discussed above with reference to Fig. 16. Although the operations of block 1800 have been described with respect to the new individual, the person of ordinary skill will readily 25 comprehend that they can be repeated, as necessary, for existing members of the group. [00110] In block 1810, one or more pairwise functions f(Pk, PO 1 ) and/or f(C;, Cj) of principal and/or complex indicators for the new individual and the ift existing group member of the group are computed. Each of the functions may correspond to a particular area of potential conflict between the new individual and an existing group member. In 30 block 1820, the one or more pairwise functions f(Pk, Pki) and/or f(Cj, Cj 1 ) are used to compute a pairwise conflict indicator, S, for the two individuals. The models used to WO 2013/147932 PCT/US2012/064178 51 compute the functions and the pairwise conflict indicator may be determined, for example, as discussed above with reference to Fig. 16. [00111] In some embodiments, the pairwise conflict indicator for two individuals may comprise a set of z-scores {Zql, each corresponding to one or more pairwise functions of 5 principal and/or complex indicators for the two individuals. In some embodiments, the indicator may comprise one or more z-scores corresponding to one or more functions of individual principal and/or complex indicators, such as described above. A set of one or more pairwise functions may be selected, as in Fig. 16, such that the corresponding Zq corresponds to a specific area of potential conflict between the roles and/or relationship of 10 the two individuals. For example, pairwise functions may be selected such that certain z scores emphasize differences between characteristics of the two individuals, while other pairwise functions may be selected such that other z-scores emphasize similarities between other characteristics. [00112] In block 1830, it is determined whether any other group members need to be 15 analyzed for predicting suitability with the prospective new member. If so, the method returns to block 1810. Otherwise, if all members have been analyzed, the method proceeds to block 1840 where the suitability of the new individual for membership in the existing group is determined based on the set of pairwise conflict indicators {S;i corresponding to each of the existing group members. This may be done in various ways 20 depending on the particular embodiment. In some embodiments, block 1840 may comprise comparing each of the pairwise conflict indicators, Si, to a suitability threshold, Sx, and determining that the new individual is suitable based on some portion of the comparisons having the desired relationship (e.g., greater than). The portion may be all, a majority, a particular number, a particular subset, etc. Put another way, in such 25 embodiments, the suitability of the new individual may be determined based on their pairwise suitability with each of the existing group members. [00113] In other embodiments, particularly ones in which each pairwise conflict indicator Si comprises a set of z-scores {Zqi}, block 1840 may comprise computing a function of the collection of z-scores comprising the set of pairwise conflict 30 indicators{S;}. For example, if higher z-scores indicate greater likelihood of conflict, the collection of z-scores may be added to obtain an overall prediction of likelihood of WO 2013/147932 PCT/US2012/064178 52 conflict between the new member and existing group members. The computed function may then be compared against a threshold to determine overall suitability as a member of the group. [00114] In other embodiments, suitability of the new individual may be determined 5 without constraining the roles of existing group members to remain the same. In other words, the suitability of the new individual may be determined assuming that their presence will cause the roles of the existing group members to change (e.g., people move to roles in which they are more/most suitable). In this embodiment, the suitability of the individual may be determined based on whether their presence will make the group more 10 suitable for the specific task, purpose, or relationship. Accordingly, a group suitability indicator for the specific task may be determined and used, for example, in the manner described above with respect to other embodiments related to group suitability. [00115] In other embodiments, suitability of one or more individuals with respect to a group may be determined based on whether their presence in the group is likely to 15 negatively impact the group's performance, e.g., by the individual filling a negative role. For example, this embodiment may be applied to an existing group by computing a group suitability indicator S for the current existing group, computing additional group suitability indicators Si, i = 1 to n, for the existing group minus each of the n members, then comparing each Si to S to see if the deletion of a particular member causes the group 20 suitability to increase. Similarly, this embodiment may be used to a prospective member of an existing group by computing a suitability indicator for the current existing group, computing a suitability indicator for the group including the prospective member, then comparing the two suitability indicators to see if the new member is likely to negatively impact the existing group's performance. 25 [00116] Other embodiments may be used to determine strategies for a group of individuals, as illustrated by the flowchart of Fig. 19. In some embodiments, the method shown in Fig. 19 may be used to determine strategies for a group of individuals based on facial descriptor measurements for individual members of the group. In other embodiments, the method of Fig. 19 may be used to determine a strategy for dealing with 30 a group of humans in a common situation, e.g., a group of individuals in a workplace, a sports team, etc. Although Fig. 19 illustrates the one or more embodiments by blocks WO 2013/147932 PCT/US2012/064178 53 arranged in a specific order, this order is merely exemplary and the steps or operations comprising the method may be performed in a different order than shown in the figure. Moreover, a person of ordinary skill will understand that the blocks shown in Fig. 19 may be combined and/or divided into blocks having different functionality. 5 [00117] In block 1900, one or more optimal combinations of facial descriptor measurements are selected for predicting one or more characteristics of interest, such as principal indicators, complex indicator, and/or suitability indicators. Block 1900 may be carried out according to the method described above with reference to Figs. 11 or 13, or in other ways within the scope of the present embodiment. The one or more 10 characteristics of interest preferably are related to the strategies under consideration. As discussed above, block 1900 may include generation of a sample library comprising digital images, additional data, and a plurality of facial descriptor measurements. The facial measurements may be - but are not necessarily - similar to the ones shown in and discussed above in reference to Figs. 1 through 7. Non-facial physical measurements 15 may also be used with the method. [00118] In block 1902, digital images and additional data for an individual of interest are obtained and stored, in the same manner as described above with reference to the sample library (e.g., block 1100 of Fig. 11). In block 1904, reference points consistent with those added to the images in the sample library are added to the images of the 20 individual of interest. In block 1906, facial descriptor measurements consistent with those in the sample library are calculated for the individual of interest. In block 1908, the one or more optimal combinations determined in block 1900 are applied to the facial descriptor measurements of the individual to predict the one or more characteristics of interest, in a manner such as described above with reference to Figs. 12 and 14. The 25 predictors may include primary indicators, complex indicators, and/or suitability indicators. Blocks 1902 through 1908 are repeated for each individual of interest. In block 1912, the predicted characteristics of interest for the individuals are used to predict one or more characteristics for the group of individuals. Finally, in block 1914, the predicted group characteristics and/or predicted individual characteristics are used to 30 determine a strategy for the group. Other factors and information may be used in either WO 2013/147932 PCT/US2012/064178 54 or both of blocks 1912 and 1914, including factors related to individuals, factors related to the group, and/or factors related to the individual or group environment. [00119] By further example, a group characteristic such as aggression level can be predicted according to the method illustrated in Fig. 19 as follows. First, the optimal 5 combination of facial descriptor measurements determined in block 1900 can be utilized according to blocks 1902 through 1910 to predict the aggressiveness of individuals in the group. Next, the set of predicted aggression levels can be combined optimally to form a predictor, or principal indicator, of the level of aggression within the group. In other words, the principal indicator of aggression level for the ith individual in the group can 10 be expressed as Pi = al -F, + a2-F 2 + . + am- Fm, where F 1 ... Fm are the selected facial descriptor measurements and a, ... am are the optimal linear combining coefficients. Likewise, if the group comprises N individuals, the complex indicator of group aggression level can be expressed as CG = p1 P 1 + 02' P 2 + .+ ON' PN, where Pf ... pN are the optimal linear combining coefficients for the group. Furthermore, in the 15 combined group model, the individuals may be ordered in a variety of ways. For example, the principal indicators of individual aggression levels P ... PN may be rearranged in descending numerical order P', ... P'N, with P' 1 corresponding to the most aggressive individual. In other embodiments, the principal indicators of individual aggression levels P 1 ... PN may be rearranged based on the ordered values of another 20 principal indicator (e.g., dominance). In any case, each optimal coefficient Pi may be selected, at least in part, based on the position within the ordered group. This would allow, for example, coefficients Pi to be selected to emphasize the characteristics of the most aggressive and/or submissive individuals in the group. [00120] Group dynamics and group characteristics among a group of individuals can be 25 predicted in various other ways known to persons skilled in the art. For example, methods may be employed that are analogous to the Hartree-Fock equation, which is commonly used to predict the behavior of the electrons in an atom. Alternately, a computational method based on a NETLOGO model can be used. An advantage of this approach is that factors other than the behavioral traits predicted from the facial 30 descriptor measurements can be incorporated. Such factors may include the estimated WO 2013/147932 PCT/US2012/064178 55 activity level and the resulting number of interaction between individuals in a group, as well as environmental factors. [00121] Furthermore, computational models based on vector spaces may be employed in any of the above methods related to predicting characteristics of individuals, pairs of 5 individuals, or groups of individuals. Conceptually, such a computational model would be based on assigning a measurement (or combination of measurements, such as an indicator described above) to an axis (or vector) in a multi-dimensional space and determining a point in this space as an "optimum." In other words, such embodiments would identify the continuous range of values for each of the facial measurements (or 10 combinations of facial measurements) and the specific value on each that is the optimum for a specific characteristic of interest. Subsequently, a performance or suitability measure could be determined for one or more individuals (or pairs or groups) of interest based on the distance (e.g., Euclidean distance) between the values of the axis variables for the individuals of interest and the corresponding optimum value previously 15 determined. In such cases, a larger distance generally indicates lower performance or suitability of the individual of interest. [00122] The various computational methods illustrated by exemplary Figs. 11 through 19 have been described herein as generating predictors of one or more characteristic, with each predictor comprising a single value such as a number, a percentile, a z-score, etc. 20 However, persons of ordinary skill will understand that no model is capable of perfectly predicting a characteristic of a specific individual, especially if the specific individual was not part of the data set from which the model was derived. Consequently, any of the prediction models described herein may also comprise prediction intervals used to show a range of values surrounding the predicted value. The width of the prediction intervals for 25 a particular predictor depends on both the variability of the data underlying the model and the degree of confidence required in the prediction (e.g., 95% confidence level). In some embodiments, the confidence level for a predictor of a particular characteristic may be determined, for example, based on the facial measures and environmental factors related to that characteristic. For example, predictors that comprised measurements of 30 facial features correlating with capacities for self-reflection, learning, and/or flexibility WO 2013/147932 PCT/US2012/064178 56 would require larger intervals due to the effect of these traits on the actual characteristic being predicted. [00123] Persons of ordinary skill in the art would understand that any of these computational methods may be embodied in various combinations of hardware and 5 software. For instance, the computations may be carried out by a specialized or general purpose digital computer, such as a laptop, desktop, tablet, smartphone, workstation, etc. Moreover, this hardware may be programmed to carry out such computations in various ways, such as by programs written in human-readable languages such as C, C++, etc. and compiled into machine-readable code for execution. Alternately, the methods may be 10 expressed in the particular language of a specialized computational software package, such as Matlab, which are further interpreted and/or compiled into machine-readable code. [00124] For example, Fig. 9 shows various hardware and software embodiments of a system according to embodiments of the present disclosure. Various hardware devices 15 such as digital camera 904, smartphone 906, cellular phone 908, tablet computer 910, and laptop computer 912 - may be used to capture a digital image of an individual of interest, such as the exemplary horse 902. The image-capturing device may then store the image in a memory operably connected to a digital computer. This may be carried out in various ways as illustrated in Fig. 9. For example, the image capturing device may 20 transmit the image through network 916 via wireless access point 914 to digital computer 920, which may be a desktop computer, server, workstation, or the like. Moreover, wireless access point 914 may be a cellular base station, wireless LAN access point, Bluetooth access point, or any other wireless connection known to persons of ordinary skill in the art. Likewise, network 916 may be a local- or wide-area, public or private 25 network, or any combination thereof, including an intranet and/or the Internet. [00125] In other embodiments, the image capturing device transfers the captured digital image to digital computer 920 through a wired connection 922, such as a Universal Serial Bus (USB) connection. In yet other embodiments, the captured image(s) may be transferred by removing a memory card from the image capturing device and inserting it 30 into memory card reader 924 of digital computer 920, which may copy the captured images to other memory accessible by or operably connected to digital computer 920.

WO 2013/147932 PCT/US2012/064178 57 Also within the spirit and scope of the present disclosure, the image capturing device may transfer the image, via methods described above or otherwise well known in the art, to devices other than digital computer 920, such as tablet computer 910. In such embodiments, further processing according to the methods describe above will occur, for 5 example, in tablet computer 910 rather than in digital computer 920. Similarly, the image capturing device may transfer the image to network storage unit 926 that is accessible via network 916, e.g., cloud storage. Network storage unit 926 may be configured to be accessible by some or all of the other devices shown in Fig. 9. [00126] In other embodiments, further processing according to the methods described 10 above also may take place in the image capturing device itself. For example, tablet computer 910 may be used to capture images of individuals of interest, store the images in memory accessible or operably connected to it (including, for example, network storage unit 926), and then execute one or more software applications embodying one or more methods described above. Specific measurements or processed data from the image 15 capturing device may also be communicated to a central computer or central location. [00127] Moreover, the terms and descriptions used herein are set forth by way of illustration only and are not meant as limitations. Those skilled in the art will recognize that many variations are possible within the spirit and scope of the disclosure as defined in the following claims, and their equivalents, in which all terms are to be understood in 20 their broadest possible sense unless otherwise indicated.

Claims (99)

1. A method for predicting one or more characteristics of an individual, comprising: for each of a plurality of individuals of a type, storing one or more digital images representing the individual in a memory operably connected to a digital computer; annotating the one or more digital images with a plurality of reference points; associating a data set about the individual with the one or more digital images representing the individual; computing, with the digital computer, one or more metrics using the plurality of reference points; and determining one or more predictors for the one or more characteristic of the individual based on the one or more metrics.

2. The method of claim 1, wherein determining one or more predictors for the one or more characteristics of the individual comprises: determining, for each of the one or more metrics, one or more relationships using the particular metric for all of the plurality of individuals and the data sets about all of the plurality of individuals; and selecting a combination of the one or metrics as each of the one or more predictors, based on the one or more relationships.

3. The method of claim 1 wherein the individual is one of a human, a non-human mammal, and a non-mammal animal.

4. The method of claim 1, wherein the individual is a human and the one or more metrics comprise at least one metric selected from a measurement group comprising: Eye Height Proportion (EHP); Eye Depth Proportion (EDP); Maximal Eye Point Proportion (MaxEPP); Minimal Eye Point Proportion (MinEPP); WO 2013/147932 PCT/US2012/064178 59 Eye Height to Length Proportion (EHLP); Eye Extrema Angle (EEA); Upper Lateral Eye Roundness Proportion (ULERP); Lower Lateral Eye Roundness Proportion (LLERP); Upper Rostral Eye Roundness Proportion (URERP); Lower Rostral Eye Roundness Proportion (LRERP); Pupilary Area Ratio (PAR); Eye Set Angle (ESA); Eye Width Proportion (EWP); Iris Boundary Proportion (IBP); Caruncle Length Proportion (CLP); Caruncle Angular Proportion (CAP); Upper Eye Angle (UEA); Lower Eye Angle (LEA); Front Eye Angle (FEA); Eye Elevation Angle (EEA); Eye Depression Angle (EDA); Eye Set Depth Proportion (ESDP); Maxillary Palate Angle (MPA); Maxillary Plate Prominence Proportion (MPPP); Facial Length Proportion (FLP); Mid-Face Proportion (MFP); Mandible Width Proportion (MWP); Labio-Orbital Angle (LOA); Naso-Orbital Angle (NOA); Naso-Chilial Angle (NCA); Upper Lip Protuberance Proportion (ULPP); Lower Lip Protuberance Proportion (LLPP); WO 2013/147932 PCT/US2012/064178 60 Upper Lip Thickness Proportion (ULTP); Lower Lip Thickness Proportion (LLTP); Lower Lip Maxima Point Proportion (LLMPP); Upper Lip Maxima Point Proportion (ULMPP); Lower Lip Angle (LLA); Upper Lip Angle (ULA); Lower Lip Roundness Proportion (LLRP); Upper Lip Roundness Proportion (ULRP); Lip Convergence Angle (LCA); Mouth-to-Pupil Width Proportion (MtPWP); Mouth-to-Nostril Width Proportion (MtNWP); Mouth Thickness Proportion (MTP); Mouth Width Angle (MWA); Upper Lip Length Proportion (ULLP); Lower Lip Length Proportion (LLLP); Upper Labial Thickness Proportion (ULTP); Lower Labial Thickness Proportion (LLTP); Jowl Fullness Proportion (JFP); Jowl Connective Proportion (JCP); Jowl Thickness Angle (JTA); Jowl Set Angle (JSA); Lambiomental Inflexion Proportion (LIP); Chin Roundness Proportion (CRP); Chin Height Proportion (CHP); Jowl Depth Angle (JDA); Radix Protrusion Proportion (RPP); Radix Protrusion Angle (RPA); Dorsum Rounding Proportion (DRP); WO 2013/147932 PCT/US2012/064178 61 Dorsum Maximal Protuberance Point Proportion (DMPPP); Nose Rounding Proportion (NRP); Nostril Tip Rounding Proportion (NTRP); Nasal Tip Angle (NTA); Nostril Height Proportion (NHP); Nares Protrusion Proportion (NaPP); Nares Length Proportion (NaLP); Nares Inflexion Proportion (NIP); Full Nose Angle (FNA); Nasal Spine Deviation Proportion (NSDP); Nasal Tip Inflection Proportion (NTIP); Alar Flare Proportion (AFP); Nasal Bridge Width Proportion (NBWP); Nasal Spine Angle (NSA); Brow Fullness Proportion (BFP); Forehead Inflexion Proportion (FIP); Forehead Slope Proportion (FSP); Forehead Roundness Proportion (FRP); Forehead Maxima Point Proportion (FMPP); Forehead Height Proportion (FHP); Eyebrow Height Proportion (EHP); Forehead Width Proportion (FWP); Forehead Fullness Proportion (FFP); Hairline Inflection Angle (HIP); Auricle Inclination Angle (AIA); Auricle Set Angle (ASA); Ear-to-Eyebrow Height Proportion (EtEHP); Ear-to-Eye Height Proportion (EtEHP2); WO 2013/147932 PCT/US2012/064178 62 Lobe Attachment Proportion (LAP); Lobe Length Proportion (LLP); Intertragical Notch Depth Proportion (INDP); Tragus Protuberance Proportion (TPP); Tragus Mid-Point Proportion (TMPP); Antitragus Protuberance Proportion (APP); Antitragus Mid-Point Proportion (AMPP); Concha Inflexion Proportion (CIP); Upper Concha Separation Angle (UCSA); Ear Tip Length Proportion (ETLP); Upper Concha Separation Angle (UCSA); Ear Roundness Proportion (ERP); Crus Helix Angle (CHA); Tubercle Proportion (TP); Helix Thickness Proportion (HTP); Lobule Inflexion Proportion (LIP);

5. The method of claim 4, wherein the one or more metrics comprise at least two metrics selected from the measurement group.

6. The method of claim 1, wherein each of the one or more metrics is one of the following types: absolute distance, normalized distance, angle, area, absolute volume, volumetric ratio, and solid angle.

7. The method of claim 1, wherein the annotating step comprises entering using a graphical user interface.

8. The method of claim 1, wherein the annotating step comprises populating automatically based on a stored profile relating to the type of the individual.

9. The method of claim 1, wherein the one or more characteristics are one or more of: temperament, cognitive ability; performance; suitability for a particular task, WO 2013/147932 PCT/US2012/064178 63 purpose, or relationship; aggressiveness; responsiveness to aggressive behavior; sensitivity; introvertedness; extrovertedness; and empathy.

10. The method of claim 1, wherein at least one of the one or more digital images is a three-dimensional image.

11. The method of claim 1, wherein determining one or more predictors for the one or more characteristics of the individual comprises determining one or more indicators used for predicting the one or more characteristics of the individual.

12. The method of claim 11, wherein the one or more indicators comprise one or more principal indicators, and wherein determining one or more indicators comprises: determining, for each of the one or more metrics, at least one principal relationship using the particular metric for all of the plurality of individuals and a primary data subset comprising the data sets about all of the plurality of individuals; selecting at least one of the one or more metrics based on the determined principal relationships; determining one or more principal functions based on the determined principal relationships; and determining each of the one or more principal indicators by combining the selected metrics combined according to one of the determined principal functions.

13. The method of claim 12, wherein the one or more principal indicators are used to predict the one or more characteristics.

14. The method of claim 12, wherein the one or more indicators comprise one or more suitability indicators used to predict the one or more characteristics, and further comprising: selecting at least one item from the set comprising the one or more metrics and the one or more principal indicators; and determining one or more suitability functions; and determining the one or more suitability indicators by combining the selected items according to one of the determined suitability functions. WO 2013/147932 PCT/US2012/064178 64

15. The method of claim 14, further comprising: determining, for each of the one or more principal indicators, at least one categorical relationship using the particular principal indicator for all of the plurality of individuals and a categorical data subset comprising the data sets about all of the plurality of individuals; and determining, for each of the one or more metrics, at least one categorical relationship using the particular metric for all of the plurality of individuals and the categorical data subset comprising the data sets about all of the plurality of individuals.

16. The method of claim 18, wherein the one or more suitability functions are determined based on the determined categorical relationships.

17. The method of claim 18, wherein the at least one item is selected from the set based on the determined categorical relationships.

18. The method of claim 12, wherein the one or more indicators comprise one or more secondary indicators, and wherein determining one or more indicators further comprises: determining, for each of the one or more principal indicators, at least one secondary relationship using the particular principal indicator for all of the plurality of individuals and a secondary data subset comprising the data sets about all of the plurality of individuals; determining, for each of the one or more metrics, at least one secondary relationship using the particular metric for all of the plurality of individuals and the secondary data subset comprising the data sets about all of the plurality of individuals; selecting at least one item from the set comprising the one or more metrics and the one or more principal indicators, based on the determined secondary relationships; and determining one or more secondary functions based on the determined secondary relationships, determining each of the one or more secondary indicators by combining the selected items combined according to one of the determined secondary functions.

19. The method of claim 18, wherein the one or more secondary indicators are used to predict the one or more characteristics. WO 2013/147932 PCT/US2012/064178 65

20. The method of claim 18, wherein the one or more indicators comprise one or more suitability indicators used to predict the one or more characteristics, and further comprising: selecting at least one item from the set comprising the one or more metrics, the one or more principal indicators, and the one or more secondary indicators; determining one or more suitability functions; and determining the one or more suitability indicators by combining the selected items according to the one of the determined suitability functions.

21. The method of claim 20, further comprising: determining, for each of the one or more secondary indicators, at least one categorical relationship using the particular secondary indicator for all of the plurality of individuals and a categorical data subset comprising the data sets about all of the plurality of individuals; determining, for each of the one or more principal indicators, at least one categorical relationship using the particular principal indicator for all of the plurality of individuals and a categorical data subset comprising the data sets about all of the plurality of individuals; and determining, for each of the one or more metrics, at least one categorical relationship using the particular metric for all of the plurality of individuals and the categorical data subset comprising the data sets about all of the plurality of individuals.

22. The method of claim 21, wherein the one or more suitability functions are determined based on the determined categorical relationships.

23. The method of claim 21, wherein the at least one item is selected from the set based on the determined categorical relationships.

24. A method for predicting one or more characteristics of an individual based on one or more metrics related to the individual, comprising: storing one or more digital images representing the individual in a memory operably connected to a digital computer, in association with digital images representing a plurality of other individuals; annotating the one or more digital images with a plurality of reference points; WO 2013/147932 PCT/US2012/064178 66 computing, with the digital computer, the one or more metrics using the plurality of reference points; and predicting the one or more characteristics based on the one or more metrics.

25. The method of claim 24, wherein the predicting step further comprises: computing a combined metric based on a predetermined function of the one or more metrics; and predicting the one or more characteristics based on the combined metric.

26. The method of claim 25, wherein the combined metric is calculated using a non-linear combination of the one or more individual metrics.

27. The method of claim 26, wherein the non-linear combination is determined using one of Newton's Method and Lagrange's Method.

28. The method of claim 25, wherein the predetermined function is determined from one or more metrics relating to a plurality of other individuals of the same type as the individual.

29. The method of claim 24, wherein the one or more characteristics relates to one or more of: the type of task most suitable for the individual and the expected performance of the individual in a particular task.

30. The method of claim 24, wherein the predicting step is performed using a computer.

31. The method of claim 24, wherein the individual is one of a human, a non human mammal, and a non-mammal animal.

32. The method of claim 24, wherein predicting the one or more characteristics is further based on at least one of additional information related to the individual and information related to the individual's environment.

33. The method of claim 24, wherein predicting the one or more characteristics of the individual based on the one or more metrics comprises computing one or more indicators used for predicting the one or more characteristics. WO 2013/147932 PCT/US2012/064178 67

34. The method of claim 33, wherein computing one or more indicators used for predicting the one or more characteristics comprises computing one or more principal indicators based on one or more predetermined functions of the one or more metrics.

35. The method of claim 34, wherein the one or more principal indicators are used to predict the one or more characteristics.

36. The method of claim 34, further comprising: computing values for one or more suitability functions based on a predetermined function of a predetermined set selected from among the one or more principal indicators and the one or more metrics; predicting the one or more characteristics based on the computed values.

37. The method of claim 34, wherein computing one or more indicators used for predicting the one or more characteristics further comprises computing one or more secondary indicators based on one or more predetermined functions of one or more predetermined sets selected from among the one or more principal indicators and the one or more metrics.

38. The method of claim 37, wherein the one or more secondary indicators are used to predict the one or more characteristics.

39. The method of claim 37, further comprising: computing values for one or more suitability functions based on a predetermined function of a predetermined set selected from among the one or more secondary indicators, the one or more principal indicators, and the one or more metrics; predicting the one or more characteristics based on the computed values.

40. The method of claim 33, wherein predicting the one or more characteristics of the individual based on the one or more metrics further comprises comparing each of the one or more indicators to a threshold.

41. The method of claim 33, wherein the one or more indicators comprise one or more of the following: a z-score, a raw score, and a percentile. WO 2013/147932 PCT/US2012/064178 68

42. A method for predicting one or more characteristic of a pair of individuals, comprising: for each individual of a plurality of pairs of individuals, storing one or more digital images representing the individual in a memory operably connected to a digital computer; annotating the one or more digital images with a plurality of reference points; associating a data set about the individual with the one or more digital images representing the individual; computing, with the digital computer, one or more metrics using the plurality of reference points; for each of the plurality of pairs of individuals, associating a pairwise data set about the pair with the one or more digital images representing the individuals comprising the pair; determining one or more indicators used for predicting the one or more characteristics of the pair.

43. The method of claim 42, wherein the one or more indicators comprise one or more principal indicators, and wherein determining one or more indicators comprises: determining, for each of the one or more metrics, at least one principal relationship using the particular metric computed for all of the individuals comprising the plurality of pairs and a primary data subset comprising the data sets about all of the individuals comprising the plurality of pairs; selecting at least one of the one or more metrics based on the determined principal relationships; determining one or more principal functions based on the determined principal relationships; and determining each of the one or more principal indicators by combining the selected metrics combined according to one of the determined principal functions.

44. The method of claim 43, wherein determining one or more indicators comprises further comprises: WO 2013/147932 PCT/US2012/064178 69 determining, for each of the one or more principal indicators, at least one categorical relationship using a plurality of pairs of the particular principal indicator and the pairwise data sets about all of the plurality of pairs of individuals, wherein each of the plurality of pairs of the particular principal indicator corresponds to one of the plurality of pairs of individuals; determining, for each of the one or more metrics, at least one categorical relationship using a plurality of pairs of the particular metric and the pairwise data sets about all of the plurality of pairs of individuals, wherein each of the plurality of pairs of the particular metric corresponds to one of the plurality of pairs of individuals; selecting at least one item from the set comprising the one or more metrics and the one or more principal indicators; and determining at least one combination functions for each of the selected items, wherein the combination function determined for each selected item comprises a pair of instances of the selected item, the pair of instances corresponding to individuals comprising a pair.

45. The method of claim 44, wherein the one or more combination functions are determined based on the determined categorical relationships.

46. The method of claim 44, wherein the at least one item is selected from the set based on the determined categorical relationships.

47. The method of claim 44, wherein the one or more combination functions are used to predict the one or more characteristics of the pair.

48. The method of claim 44, wherein the one or more indicators comprise one or more suitability indicators used to predict the one or more characteristics of the pair, and further comprising: selecting at least one item from the set comprising the combination functions, the one or more principal indicators, and the one or more metrics; determining one or more suitability functions; and determining the one or more suitability indicators by combining the selected items according to the one of the determined suitability functions. WO 2013/147932 PCT/US2012/064178 70

49. The method of claim 48, wherein the one or more suitability functions are determined based on one or more of the determined categorical relationships and the determined principal relationships.

50. The method of claim 48, wherein the at least one item is selected from the set based on one or more of the determined categorical relationships and the determined principal relationships.

51. The method of claim 48, wherein the one or more indicators comprise one or more secondary indicators, and wherein determining one or more indicators further comprises: determining, for each of the one or more metrics, at least one secondary relationship using the particular metric computed for all of the individuals comprising the plurality of pairs and a secondary data subset comprising the data sets about all of the individuals comprising the plurality of pairs; determining, for each of the one or more principal indicators, at least one secondary relationship using the particular principal indicators computed for all of the individuals comprising the plurality of pairs and a secondary data subset comprising the data sets about all of the individuals comprising the plurality of pairs; selecting at least one item from the set comprising the one or more metrics and the one or more principal indicators, based on the determined secondary relationships; determining one or more secondary functions based on the determined secondary relationships; and determining each of the one or more secondary indicators by combining the selected items according to one of the determined secondary functions.

52. The method of claim 51, wherein determining one or more indicators further comprises: determining, for each of the one or more secondary indicators, at least one categorical relationship using a plurality of pairs of the particular secondary indicator and the pairwise data sets about all of the plurality of pairs of individuals, wherein each of the plurality of WO 2013/147932 PCT/US2012/064178 71 pairs of the particular secondary indicator corresponds to one of the plurality of pairs of individuals; determining, for each of the one or more principal indicators, at least one categorical relationship using a plurality of pairs of the particular principal indicator and the pairwise data sets about all of the plurality of pairs of individuals, wherein each of the plurality of pairs of the particular principal indicator corresponds to one of the plurality of pairs of individuals; determining, for each of the one or more metrics, at least one categorical relationship using a plurality of pairs of the particular metric and the pairwise data sets about all of the plurality of pairs of individuals, wherein each of the plurality of pairs of the particular metric corresponds to one of the plurality of pairs of individuals; selecting at least one item from the set comprising the one or more metrics, the one or more principal indicators, and the one or more secondary indicators; and determining at least one combination function for each of the selected items, wherein the combination function determined for each selected item comprises a pair of instances of the selected item, the pair of instances corresponding to individuals comprising a pair.

53. The method of claim 52, wherein the one or more combination functions are determined based on the determined categorical relationships.

54. The method of claim 52, wherein the at least one item is selected from the set based on the determined categorical relationships.

55. The method of claim 52, wherein the one or more combination functions are used to predict the one or more characteristics of the pair.

56. The method of claim 52, wherein the one or more indicators comprise one or more suitability indicators used to predict the one or more characteristics of the pair, and further comprising: selecting at least one item from the set comprising the combination functions, the one or more principal indicators, and the one or more metrics; determining one or more suitability functions; and WO 2013/147932 PCT/US2012/064178 72 determining the one or more suitability indicators by combining the selected items according to one of the determined suitability functions.

57. The method of claim 56, wherein the one or more suitability functions are determined based on one or more of the determined categorical relationships, the determined principal relationships, and the determined secondary relationships.

58. The method of claim 56, wherein the at least one item is selected from the set based on one or more of the determined categorical relationships, the determined principal relationships, and the determined secondary relationships.

59. The method of claim 42, wherein both individuals of a pair are members of the same species.

60. The method of claim 42, wherein both individuals of a pair are members of different species.

61. A method for predicting one or more characteristics of a pair of individuals based on one or more metrics related to the individuals of the pair, comprising: storing one or more digital images representing each of the pair of individuals in a memory operably connected to a digital computer; annotating the one or more digital images with a plurality of reference points; computing, with the digital computer, the one or more metrics using the plurality of reference points; and predicting the one or more characteristics of the pair based on the one or more metrics.

62. The method of claim 61, wherein predicting the one or more characteristics of the pair based on the one or more metrics comprises computing one or more indicators used for predicting the one or more characteristics.

63. The method of claim 62, wherein computing one or more indicators used for predicting the one or more characteristics comprises computing one or more principal indicators based on one or more predetermined functions of the one or more metrics. WO 2013/147932 PCT/US2012/064178 73

64. The method of claim 63, wherein the one or more principal indicators are used to predict the one or more characteristics.

65. The method of claim 63, further comprising: computing values for each of the one or more suitability functions based on a predetermined function of a predetermined set selected from among the one or more principal indicators and the one or more metrics; predicting the one or more characteristics based on the computed values.

66. The method of claim 65, wherein each of the predetermined functions of the predetermined sets is based on at least one pair of instances of the same item, wherein the pair of instances corresponds to individuals comprising a pair.

67. The method of claim 63, wherein computing one or more indicators used for predicting the one or more characteristics further comprises computing one or more secondary indicators based on one or more predetermined functions of one or more predetermined sets selected from among the one or more principal indicators and the one or more metrics.

68. The method of claim 67, wherein the one or more secondary indicators are used to predict the one or more characteristics.

69. The method of claim 67, further comprising: computing values for each of the one or more suitability functions based on a predetermined function of a predetermined set selected from among the one or more secondary indicators, the one or more principal indicators, and the one or more metrics; predicting the one or more characteristics based on the computed values.

70. The method of claim 69, wherein each of the predetermined functions of the predetermined sets is based on at least one pair of instances of the same item, wherein the pair of instances corresponds to individuals comprising a pair.

71. The method of claim 61, wherein predicting the one or more characteristics of the individual based on the one or more metrics further comprises comparing each of the one or more indicators to a threshold. WO 2013/147932 PCT/US2012/064178 74

72. The method of claim 62, wherein the one or more indicators comprise one or more of the following: a z-score, a raw score, and a percentile.

73. The method of claim 61, wherein both individuals of a pair are members of the same species.

74. The method of claim 61, wherein both individuals of a pair are members of different species.

75. A method for determining the compatibility of an individual with other individuals comprising a group, comprising: storing one or more digital images representing the individual in a memory operably connected to a digital computer; annotating the one or more digital images with a plurality of reference points; computing, with the digital computer, one or more metrics related to the individual using the plurality of reference points; and predicting the compatibility of the individual based on the one or more metrics related to the individual and indicators of characteristics related to the other individuals.

76. The method of claim 75, wherein predicting the compatibility of the individual comprises: computing one or more individual indicators related to characteristics of the individual based on one or more predetermined functions of the one or more metrics; obtaining one or more individual indicators related to the same characteristics of the other individuals comprising the group; for each of the other individuals comprising the group: computing one or more pairwise indicators related the individual and the particular other individual, each pairwise indicator computed based on a predetermined function of a predetermined set of indicators comprising at least one indicator related to the individual and the same at least one indicator related the particular other individual of the group; WO 2013/147932 PCT/US2012/064178 75 computing a pairwise suitability indicator based on a predetermined function of the one or more pairwise indicators related to the individual and the particular other individual; and predicting the compatibility of the individual based on a predetermined function of the respective pairwise suitability indicators related to the individual and each of the other individuals comprising the group.

77. The method of claim 76, wherein at least one of the pairwise suitability indicators comprises at least one z-score.

78. The method of claim 77, wherein each of the pairwise suitability indicators comprises at least one z-score, and the predetermined function of the respective pairwise suitability indicators comprises a sum of the z-scores comprising the pairwise suitability indicators.

79. The method of claim 76, where the predetermined function of the respective pairwise suitability indicators comprises comparing each of the pairwise suitability indicators to a respective suitability threshold value.

80. A method for predicting one or more characteristics of a group of individuals of the same type, comprising: for each of the individuals comprising the group, storing one or more digital images representing the individual in a memory operably connected to a digital computer; annotating the one or more digital images with a plurality of reference points; computing, with the digital computer, one or more metrics using the plurality of reference points; predicting one or more characteristics of the individual based on the one or more metrics; and predicting the one or more characteristics of the group of individuals based on the predicted one or more characteristics of the individuals comprising the group. WO 2013/147932 PCT/US2012/064178 76

81. The method of claim 80, further comprising determining a strategy related to the group of individuals based on the predicted one or more characteristics of the group of individuals.

82. The method of claim 80, wherein predicting one or more characteristics of the individual further comprises: computing one or more combined metrics, each based on a predetermined function of the one or more metrics; and predicting the one or more characteristics of the individual based on the one or more combined metrics.

83. The method of claim 82, wherein the predetermined function is determined from the one or more metrics related to a plurality of individuals of the same type as the individuals comprising the group.

84. The method of claim 80, wherein predicting one or more characteristics of the group of individuals further comprises: computing one or more group combined metrics, each based on a predetermined function of the predicted one or more characteristics of the individuals comprising the group; and predicting the one or more characteristics of the group of individuals based on the one or more group combined metrics.

85. The method of claim 80, wherein predicting one or more characteristics of the group of individuals further comprises: for each of the one or more metrics, computing an average for the individuals comprising the group; and computing one or more group combined metrics, each based on a predetermined function of the computed averages of the one or more metrics; and predicting the one or more characteristics of the group of individuals based on the one or more group combined metrics. WO 2013/147932 PCT/US2012/064178 77

86. The method of claim 80 wherein predicting the one or more characteristics of the group of individuals is further based on at least one of estimated number of interactions between the individuals comprising the group; and environmental information related to the group of individuals.

87. A method for predicting a characteristic relating to a type of individual for an individual of the type, comprising: measuring, using a computer, two or more metrics from reference points on a digital image representing the individual of the type; calculating, using a computer, a resultant using the two or more measured metrics and a mathematical formula; predicting, using a computer, a behavioral or performance characteristic of the individual using the resultant from the calculated mathematical formula for the individual and data about a group individuals of the same type as the individual; and presenting the predicted behavioral or performance characteristic of the individual.

88. The method of claim 87 wherein the data about the group of individuals is calculated using the mathematical formula.

89. The method of claim 88 further comprising: obtaining the digital image representing the individual ; and assigning the reference points on the obtained digital image.

90. A system for predicting one or more characteristics of an individual, comprising: at least one processor; at least one memory, operably connected to the at least one processor and configured to store data and software code that, when executed by the at least one processor, causes the system to: for each of a plurality of individuals of a type, store one or more digital images representing the individual in the at least one memory; WO 2013/147932 PCT/US2012/064178 78 annotate the one or more digital images with a plurality of reference points; associate a data set about the individual with the one or more digital images representing the individual; compute one or more metrics using the plurality of reference points; and determine one or more predictors for the one or more characteristic of the individual based on the one or more metrics.

91. A computer readable medium comprising a set of instructions that, when executed by a processor comprising a computing device, causes the computing device to: for each of a plurality of individuals of a type, store one or more digital images representing the individual in a memory operably connected to the processor; annotate the one or more digital images with a plurality of reference points; associate a data set about the individual with the one or more digital images representing the individual; compute one or more metrics using the plurality of reference points; and determine one or more predictors for one or more characteristics of individuals of the type based on the one or more metrics.

92. A system for predicting one or more characteristics of an individual based on one or more metrics related to the individual, comprising: at least one processor; at least one memory, operably connected to the at least one processor and configured to store data and software code that, when executed by the at least one processor, causes the system to: store one or more digital images representing the individual in the at least one memory, in association with digital images representing a plurality of other individuals; WO 2013/147932 PCT/US2012/064178 79 annotate the one or more digital images with a plurality of reference points; compute the one or more metrics using the plurality of reference points; and predict the one or more characteristics based on the one or more metrics and one or more predetermined functions.

93. A computer readable medium comprising a set of instructions that, when executed by a processor comprising a computing device, causes the computing device to: store one or more digital images representing an individual in a memory operably connected to the processor, in association with digital images representing a plurality of other individuals; annotate the one or more digital images with a plurality of reference points; compute the one or more metrics using the plurality of reference points; and predict the one or more characteristics of the individual based on the one or more metrics and one or more predetermined functions.

94. A system for predicting one or more characteristics of a pair of individuals, comprising: at least one processor; at least one memory, operably connected to the at least one processor and configured to store data and software code that, when executed by the at least one processor, causes the system to: for each of a plurality of individuals of a type, store one or more digital images representing the individual in the at least one memory; annotate the one or more digital images with a plurality of reference points; associate a data set about the individual with the one or more digital images representing the individual; compute one or more metrics using the plurality of reference points; and WO 2013/147932 PCT/US2012/064178 80 for one or more of the pairs of individuals, associate a pairwise data set about the pair with the one or more digital images representing the individuals comprising the pair; determine one or more indicators used for predicting the one or more characteristics of the pair based on the one or more metrics.

95. A computer readable medium comprising a set of instructions that, when executed by a processor comprising a computing device, causes the computing device to: for each individual of a plurality of pairs of individuals of a type, store one or more digital images representing the individual in a memory operably connected to the processor; annotate the one or more digital images with a plurality of reference points; associate a data set about the individual with the one or more digital images representing the individual; compute one or more metrics using the plurality of reference points; and for one or more of the pairs of individuals, associate a pairwise data set about the pair with the one or more digital images representing the individuals comprising said pair; determine one or more indicators used for predicting the one or more characteristics of the pair based on the one or more metrics.

96. A system for determining the compatibility of an individual with other individuals comprising a group, comprising: at least one processor; at least one memory, operably connected to the at least one processor and configured to store data and software code that, when executed by the at least one processor, causes the system to: store one or more digital images representing the individual in a memory operably connected to a digital computer; annotate the one or more digital images with a plurality of reference points; WO 2013/147932 PCT/US2012/064178 81 compute, with the digital computer, one or more metrics related to the individual using the plurality of reference points; and predict the compatibility of the individual based on the one or more metrics related to the individual and indicators of characteristics related to the other individuals.

97. A computer readable medium comprising a set of instructions for determining the compatibility of an individual with a group that, when executed by a processor comprising a computing device, causes the computing device to: store one or more digital images representing an individual in a memory; annotate the one or more digital images with a plurality of reference points; compute one or more metrics related to the individual using the plurality of reference points; and predict the compatibility of the individual with other individuals in a group based on the one or more metrics related to the individual and indicators of characteristics related to the other individuals in the group.

98. A system for predicting one or more characteristics of a group of individuals of the same type, comprising: at least one processor; at least one memory, operably connected to the at least one processor and configured to store data and software code that, when executed by the at least one processor, causes the system to: for each of the individuals comprising the group, store one or more digital images representing the individual in the at least one memory; annotate the one or more digital images with a plurality of reference points; compute one or more metrics using the plurality of reference points; and WO 2013/147932 PCT/US2012/064178 82 predicting one or more characteristics of the individual based on the one or more metrics; and predicting the one or more characteristics of the group of individuals based on the predicted one or more characteristics of the individuals comprising the group.

99. A computer readable medium comprising a set of instructions for predicting one or more characteristics of a group of individuals of the same type that, when executed by a processor comprising a computing device, causes the computing device to: for each of the individuals comprising the group, store one or more digital images representing the individual in the at least one memory; annotate the one or more digital images with a plurality of reference points; compute one or more metrics using the plurality of reference points; and predicting one or more characteristics of the individual based on the one or more metrics; and predicting the one or more characteristics of the group of individuals based on the predicted one or more characteristics of the individuals comprising the group.