WO2011048219A2 - Methode et systeme pour evaluer la ressemblance d'un objet requete a des objets de reference - Google Patents
Methode et systeme pour evaluer la ressemblance d'un objet requete a des objets de reference Download PDFInfo
- Publication number
- WO2011048219A2 WO2011048219A2 PCT/EP2010/065997 EP2010065997W WO2011048219A2 WO 2011048219 A2 WO2011048219 A2 WO 2011048219A2 EP 2010065997 W EP2010065997 W EP 2010065997W WO 2011048219 A2 WO2011048219 A2 WO 2011048219A2
- Authority
- WO
- WIPO (PCT)
- Prior art keywords
- data
- class
- regions
- user
- similarity
- Prior art date
Links
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
- G06F18/213—Feature extraction, e.g. by transforming the feature space; Summarisation; Mappings, e.g. subspace methods
- G06F18/2137—Feature extraction, e.g. by transforming the feature space; Summarisation; Mappings, e.g. subspace methods based on criteria of topology preservation, e.g. multidimensional scaling or self-organising maps
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
- G06F16/10—File systems; File servers
- G06F16/14—Details of searching files based on file metadata
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
- G06F16/90—Details of database functions independent of the retrieved data types
- G06F16/93—Document management systems
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F17/00—Digital computing or data processing equipment or methods, specially adapted for specific functions
- G06F17/10—Complex mathematical operations
- G06F17/16—Matrix or vector computation, e.g. matrix-matrix or matrix-vector multiplication, matrix factorization
Definitions
- the present invention provides a method and system for evaluating the similarity of a request object to reference objects. It applies for example in the field of pattern recognition.
- the object can be a material object, an individual, the state of a system, or a group of such objects, individuals or states, whose physical characteristics are measured.
- Discriminators are used to classify objects, that is, to make a decision about whether an object belongs to a class or to several predefined classes of objects. For example, the physical characteristics of a patient's condition are measured, such as height, weight, age, blood pressure, and body temperature. This patient is the object query or patient query. The measured values are then provided to a discriminator, who compares these characteristics with those of other patients identified as suffering from a particular disease. These other patients are reference objects or reference patients. Diseases are classes. The discriminator attributes to the patient the same illness as that of the closest reference patients in the sense of a proximity measure based on the characteristics measured. Such applications are used for decision support, they can even do without the opinion of the expert.
- decision support systems propose to associate a confidence index with the decision made by the discriminator.
- the proposed decision is not explicit, the system behaves like a black box, but it provides the user with a clue supposed to reassure him on the quality of this decision.
- This index can be of probabilistic nature, or obtained by the relative comparison of the decisions of several different discriminators.
- this index is obtained by a relatively complex process from the point of view of the non-specialist user of discrimination methods.
- the broader system of discrimination that provides both a decision and a confidence index for this decision is a judge and a party, which is not conducive to inspiring user confidence.
- Other decision support systems are based on the explanation of the decision made by the discriminator in intelligible terms by the user.
- fuzzy inference systems explain their decision as the result of a weighted sum of logical rules directly implicating the original characteristics of the objects, these quantities and their combination being assumed to be easy to interpret by the user.
- the number of rules and parameters is often important, which greatly reduces the intelligibility of the system.
- the projection methods induce a loss of information called false neighborhood or glue, which artificially brings artifacts closer to the extent of their similarity.
- false neighborhoods exist either because it is technically impossible to respect all the similarities during the projection, or because, although it is technically possible, the projection method could not find this solution.
- the user can assign to a request object the majority class of surrounding objects on the map, even if these objects are falsely close to the request object.
- the reference objects close to the request object in the form of a list of reference objects ordered according to their decreasing proximities to the request object.
- search engines on the Internet where a query results in the display of a list of reference internet pages, ordered by proximity to the query.
- the list of reference objects presented to the user is dependent on the query used on the one hand, and has a linear order on the other hand.
- the request results in the display of a set of reference web pages ordered in the form of groups on a flat map and highlighted graphically (color, size ...) so as to signify their proximity to the query.
- the card presented to the user depends on the request.
- the user can not build a stable mental representation of the universe of reference objects, this universe never being presented to him completely or independently of the request. He can not judge for himself the quality of the proximity information to the request, which is presented to him. Nor can it easily apprehend the similarities or differences between the query objects translated by their representation in terms of reference objects, since it does not have a fixed base of comparison.
- the object of the invention is notably to allow the user to obtain a stable mental representation of the universe of the reference objects independently of the request object.
- the invention proposes to cut the card into disjoint zones each associated with a reference object, and to indicate to the user in each zone, the degree of resemblance between the request object and the reference object of this zone.
- the subject of the invention is a method for evaluating the class of test data in a data space of dimension D where D> 3, each data item belonging to at least one class containing one or more data.
- the method includes a step of projecting a reference dataset of the data space into a Q dimension space where Q ⁇ D, the class of each reference datum being known.
- the method also includes a step of calculating a measure of similarity of the test data to each of the reference data.
- the method also includes a step of partitioning the projection space into a plurality of disjoint regions each containing the projection of one and only one reference datum.
- the method includes a step of evaluating the class of the test data, this class being evaluated as being the same class as one of the reference data contained in one of the regions containing the reference data closest to the data item. test in the sense of the similarity measure. Indeed, these regions are the regions most likely to contain a projection of the test data.
- the data may be digitized data
- the digitized data may include one or more physical feature measurements of an object, whether it is a hardware object or a group of hardware objects, or whether it is an individual or a group of individuals, or whether it is a state of a system or a group of states of a system, whose characteristics physical properties can be measured.
- the reference data can be projected into the projection space so as to minimize a function dependent on the similarity measure between the reference data and the distance between the projections of said reference data, so as to preserve, in the projection space, the spatial organization of the reference data.
- the regions may be the Voronoi regions associated with projections of the reference data in the projection space.
- the data can be digitized handwritten characters, the classes can group identical characters, each data can be defined by a pixel vector.
- the data can be digitized seismic curves, a class that can group the curves whose recording corresponds to an earthquake and another class that can group the curves whose recording does not correspond to an earthquake. Earth.
- the data can be digital photographs of melanomas, a class that can group the photographs of malignant melanomas, and another class that can group the benign melanoma photographs.
- the present invention also provides a method for assisting a user in deciding the class of test data in a data space of dimension D where D> 3, each data belonging to a class containing one or more data.
- the method comprises a step according to the invention for evaluating the class of the test data, as well as a step of presenting the user with the regions containing the projections of the reference data which are the closest to the test data. sense of the similarity measure.
- the region containing the projection of the reference datum that is closest to the test datum in the sense of the similarity measure can be presented to the user by using a predefined color to represent it.
- the regions containing the projections of the reference data that are closest to the test data in the sense of the similarity measure can be presented to the user by using predefined colors to represent them, so as to represent the regions in descending order of similarity with the test data.
- the method may comprise a step of assigning the user of a class to the test data item, the class assigned by the user to the test data item may or may not be the class of a reference datum contained. in one of the regions presented to the user.
- the present invention also relates to a pattern recognition device, characterized in that it implements a method according to the invention.
- the present invention also relates to a system for evaluating the class of a test data in a data space of dimension D where D> 3, each data belonging to at least one class containing one or more data.
- the system includes a projection module of a reference dataset of the data space in a Q dimension space where Q ⁇ D, the class of each reference datum being known.
- the system also includes a module for calculating a measure of similarity of the test data to each of the reference data.
- the system also includes a partitioning module of the projection space into a plurality of disjoint regions each containing the projection of one and only one reference datum.
- the system also includes a class evaluation module of the test data, which class is evaluated as being the same class as one of the reference data contained in one of the regions containing the reference data closest to the data item. test in the sense of the similarity measure. Indeed, these regions are the regions most likely to contain a projection of the test data.
- the data may be digitized data
- the digitized data may include one or more physical feature measurements of an object, whether it is a hardware object or a group of hardware objects, or whether it is an individual or a group of individuals, or whether it is a state of a system or a group of states of a system, whose characteristics physical properties can be measured.
- the reference data can be projected into the projection space so as to minimize a function dependent on the similarity measure between the reference data and the distance between the projections of said reference data, so as to preserve, in the projection space, the spatial organization of the reference data.
- the regions may be the Voronoi regions associated with projections of the reference data in the projection space.
- the data can be digitized handwritten characters, the classes can group identical characters, each data can be defined by a pixel vector.
- the data can be digitized seismic curves, a class that can group the curves whose recording corresponds to an earthquake and another class that can group the curves whose recording does not correspond to an earthquake. Earth.
- the data can be digital photographs of melanomas, a class that can group the photographs of malignant melanomas, and another class that can group the benign melanoma photographs.
- the present invention also provides a system for assisting a user in deciding the class of test data in a data space of dimension D where D> 3, each data belonging to a class containing one or more data.
- the system comprises a module according to the invention for evaluating the test data class and a module for presenting the user with the regions containing the projections of the reference data that are closest to the test data in the sense of the measure of similarity.
- the region containing the projection of the reference datum that is closest to the test datum in the sense of the similarity measure can be presented to the user by using a predefined color to represent it.
- the regions containing the projections of the reference data that are closest to the test data in the sense of the similarity measure can be presented to the user by using predefined colors to represent them, so as to represent the regions in descending order of similarity with the test data.
- the system may comprise a module for assigning the user of a class to the test data, the class assigned by the user to the test data which may or may not be the class of reference data contained in one of the regions presented to the user.
- the main advantages of the invention are that it provides, from a map of the reference objects, a graphical means that makes all the similarities between the request object and the reference objects immediately intelligible to the user without inducing choice a priori.
- another advantage of the invention is that it does not provide a decision: the user knows that he remains indispensable in his decision-making role, which is conducive to maintaining a sense of responsibility towards him of decision. Similarly, if he wished to evade moral or legal obligations, by using the facility of relying on a decision provided by an automatic system, he does not have that possibility with him. invention. This lack of automatic decision is also conducive to the user projecting on the method a collaborative behavior rather than competitive, likely to increase the confidence it has in it.
- a system implementing the method according to the invention can be implemented on most computers equipped with a graphic display device.
- the relative positioning of the reference objects together with their similarity measurement to the request object advantageously makes it possible to order the degree of resemblance of the request object to the reference objects, no longer only according to the measurement of the degree of similarity between the request object and reference objects, but according to both this degree of similarity and the relative similarities between reference objects themselves.
- the representation space of the reference objects obtained by the method according to the invention makes it possible to determine groups of objects of similar reference in terms of their position in the projection base, each of these groups containing reference objects resembling to a certain degree the query object in terms of the similarity measure. From these groups, it is possible, for example, to assign to the request object the most significant class in the reference object group that is the most similar to the request object, then to assign the second most the request object is the most significant class in the second group of reference objects most similar to the request object, and so for the existing G groups.
- the representation space of the reference objects obtained by the method according to the invention thus allows this more accurate estimation of the similarity of the request object to the reference objects.
- FIG. 1 bis, by a diagram, of the projection steps according to the invention
- FIG. 2 another example of a data card that can be used to implement the invention
- FIG. 1 illustrates an example of a map of animal species according to the invention.
- a request object corresponding to the human species, not shown in FIG. 1 is characterized by a set of similarity measurements to reference objects, each object corresponding to another animal species.
- one reference object corresponds to the species of clams, another to the species of lobsters, another to the species of ladybugs, another to the species of bees, another to the species of kiwis, another to the species of soles, another to the species of haddocks, another to crows, gulls, catfish, dolphins, and toads, hamsters, another for the seal species, another for the piranha species, another for the hare species, another for the mole species, another for the goat species, another for the species of pumas and another species of gorillas.
- the region around each reference object can be colored according to the proximity of the man to the corresponding species, this according to a given metric. Thanks to the invention, we can observe the proximity of
- a set S contains N reference objects, which are described by a set of NxK similarity measures ... N between each object i of S and a query object q which does not belong to S.
- the similarities z m iq are real numbers whose value is a function taking at least the argument object i considered and the object q.
- z mi q can be obtained from a measure of distance defined between the objects i and q represented as vectors of D characteristics v, and v q defined in a base B D of IR D. It may possibly miss similarity values z mi q , the absence of value is then coded in a specific way.
- the class of membership of the objects can possibly be provided, it can take the form of a value taken from C possible values, each identifying a class of membership.
- the function f iz may depend on other parameters, but in any case it depends on at least z m iq .
- the reference objects are positioned on a map, that is to say a metric space with Q dimensions defined by a base B Q , and the position of the reference object i is defined by a vector w, at Q components in B Q.
- This positioning can be natural, the reference objects can already have coordinates for representation on the map. Otherwise, this positioning can be manual, performed by the user of the invention. Or this positioning can be automatic: similarity measurements between reference objects, or their position in the base B D , are then used to define the position in the base B Q of the reference objects.
- the reference objects can be positioned on the map so as to minimize a function of the similarity measurements between the reference objects and distances between their projections on the map so as to preserve on the map the spatial organization of the objects.
- this function being for example the weighted sum of the absolute values of the two-to-two differences raised to a power x between the measurement of similarity between the reference objects and the Euclidean distance between their projections on the map , the weighting being a function of the similarity measures between the reference objects and the distances between their projections on the map, for example to favor the preservation of small distances rather than large ones, and the power x being a real number.
- each reference object i is represented by a region R, described below, whose position determines w.
- a region R is defined whose appearance is parameterized by the K real numbers , ..., p iK ) - Specific appearances are associated with the different possible combinations of no value for these parameters.
- the Voronoi region can be colored gray by the point of coordinates w, the luminous intensity of this gray being proportional to the value of a parameter p.
- the Voronoi region of the point w can be colored from the color scale Red, Green, Blue, the color being defined by the value of three parameters pn, p i2 and p i3 in this scale. Parameters p can also be used to modify the shape or the size, for example, of the regions Ri. In other words, there is a possible step of calculating the appearance characteristics of the regions Ri (size, shape, color, texture, orientation, gloss) based on all available information, namely the set S reference objects in the form of their coordinates in the base B D or in the form of their coordinates in the base B Q , the set of similarity measures M of the request object to the reference objects, and the set of functions F.
- the request object can be visualized by the appearance of the regions R 1.
- This appearance can be determined by calculating the parameters p iz which are functions of z mi q and possibly any set of additional parameters. If z mi q has no value (missing value), the parameter p iz does not provide a value and a specific appearance is then used.
- the reference objects S and the request object q not belonging to S can be described in the same base B D , from which, from a share the similarity measures M of each object i of S to the request object q defining the coordinates of q in the base B P are calculated as a function (for example the Euclidean distance) taking as argument the characteristics in B D of the The query object q and the objects of S, and on the other hand the characteristics of the objects of S in the base B Q are obtained by a projection method of these objects of S described in the base B D.
- a function for example the Euclidean distance
- the cartesian product of the base B Q and the base B P form a base B s in which each reference object S is described by its coordinates in the base B Q and by those in its base B P , in other words each object of S is described in the base B s by a set of characteristics obtained by projection its characteristics in B D , and a set of characteristics obtained by F functions for calculating its degree of similarity to the request object q.
- the query object q does not exist as a set of characteristics in the base B s , it only appears implicitly in its degree of similarity to the reference objects.
- the coordinates of each reference object in the base B s simultaneously carry its degree of similarity to the request object q in the part B P of B s and its degree of similarity to the other reference objects by its absolute position in the part B Q of B s which implicitly gives its position relative to the other reference objects in this base
- a step of evaluating the degree of similarity of the object query to the reference objects can take place by a function taking as argument the characteristics of the only reference objects in the database B s without the necessity of projecting the query object q into the base B Q of B s .
- the base B s can take a graphic form, the part B Q of B s giving for each reference object i its position on the screen, the part B P of B s giving the degree of resemblance of the request object q to this reference object i as a color or a specific appearance of a region R, positioned at the same place on the screen.
- the method according to the invention can be seen as a method for transforming the reference objects S and the request object q described in the database B D to a classifying description of the reference objects in the base B s.
- the degree of resemblance of the query object q to the reference objects S consubstanciel to their classification is expressed as such by the coordinates of the reference objects in the database.
- the result of the classification by the method according to the invention is given as such by the value of the coordinates of the reference objects and indirectly of the request object in the base B s .
- Figure 1a illustrates the steps of projection from B D to B Q and from B D to B P and the constitution of B s by the Cartesian product of B P and B Q , as well as the classification step of query object q in this base B s .
- FIG. 2 illustrates an example of a data card that can be used to implement the invention in a decision support system, for example a handwritten character recognition system.
- the objects can be thumbnails of handwritten figures of 8 x 8 pixels in grayscale. The thumbnails are separated into 10 balanced classes corresponding to the ten digits (0, 1, 2, 9).
- the objective is to find the class of a thumbnail query, that is to say to determine the digit from 0 to 9 which is represented by a matrix of 8 by 8 pixels, this figure not being known a priori .
- 300 reference thumbnails were arbitrarily selected in a public database.
- the 300 images were positioned in a plane to form a map, so that the thumbnails are grouped on the map in areas.
- Each zone can be easily delimited visually from the class of membership of the thumbnails it contains, that is to say, the handwritten number that represent the thumbnails it contains, this figure being known a priori.
- FIGS. 3, 4 and 5 illustrate the same data card as the card of FIG. 2, the card being used to evaluate according to the invention the figures represented by three examples of request images.
- the three thumbnails requests correspond to the thumbnails at the top left of Figures 3, 4 and 5, it is the number 0, the number 1 and the letter x respectively. It should be understood that, in accordance with the invention, these three thumbnail requests are not positioned on the map illustrated by FIGS. 3, 4 and 5.
- each thumbnail can be defined by a vector with 64 values in [0,1], each value in [0, 1] representing the luminous intensity of a pixel.
- the regions for viewing may be the Voronoi regions associated with each of the reference thumbnails.
- the Voronoi region associated with a reference thumbnail contains the entire points of the plane that are closer to the point representing this thumbnail on the map than to any other point representing a thumbnail. Then, on presentation of a request thumbnail, the Voronoi regions are colored to visualize the similarity between the object query considered and each reference object.
- This similarity is determined from the similarity measures M, which can be, in this example, the Euclidean distance in the vector space of the 64-dimensional pixels.
- the color can be all the more clear that the reference image is close to the small image request in the sense of the Euclidean distance.
- a null Euclidean distance m (identical image) can be represented by a white region
- a high Euclidean distance m can be represented by a black region.
- the minimum Euclidean distance corresponding to a black region it is for example calculated for each reference thumbnail Euclidean distance to the 6 th nearest neighbor, the minimum distance that can be the maximum of the six distances calculated.
- a Euclidean distance m greater than this minimum distance can be represented by the black color, the shorter Euclidean distances being able to be colored according to a level of color going from dark red (big m) to yellow orange, then to white (m nothing).
- the request thumbnail represents the handwritten numeral 0, as illustrated by the box at the top left of FIG. 3, a group of neighboring regions appear unambiguously very thin, especially in yellow and in orange. These are the regions associated with reference thumbnails belonging to class 0. This indicates to the user that the query thumbnail probably belongs to this class, which is indeed the case.
- the invention is therefore effective in the field of pattern recognition. In a generic application, it is an object with all the typical characteristics of a reference object. The decision is easy and the risk of error is low. As part of a decision support system, a "Nothing to report" warning signal could be issued to the user.
- the request thumbnail represents the handwritten figure 1 illustrated by the box at the top left of FIG.
- the query thumbnail represents the letter x, as illustrated by the box at the top left of Figure 5, that is to say a character that does not correspond to any reference thumbnail.
- no region is cleared, which can be interpreted as an indication of an anomaly in the data presented.
- the proposed invention easily allows the user to detect this anomaly.
- this is a new atypical object that is not identifiable to the reference classes.
- a thorough analysis is required, the creation of a new class is to consider.
- a "Stop" warning signal could be issued to the user.
- the request image is never positioned on the map, which avoids any visual contradiction between the artificial neighborhood that this positioning would induce and the real neighborhood provided by the similarity measures.
- the invention only visualizes the real neighborhood provided by the similarity measures, so as to optimize the intelligibility of the information displayed.
- a similarity measure is a function taking as argument two objects and parameters independent of these objects, such as for example a distance between two objects or an uncertainty on the distance between two objects, or a parameter d scale used to determine the dynamics of similarity measure values, such as its minimum and maximum.
- Each reference object can be characterized by its position on the map and possibly by one or more similarity measures to all the other reference objects. At least one of the reference objects can be positioned manually or automatically on the map, possibly from the set of similarity measures, so as to allow visual apprehension. In order to facilitate the visual interpretation, it is recommended a positioning such that, in the first place, the more the objects of reference are similar according to an additional measure provided, the closer they are on the map, and secondly, the objects of the same class are close and those of different classes are distant from each other on the map. But the request object is never placed on the map.
- one or more of the similarity measures between the request object and a reference object can be visualized on the map by a specific appearance, for example in terms of size, shape, color texture, or in the form of of a region associated with this reference object. This can be used to visualize the absence of measurement, or to visualize a similarity with its inaccuracy or uncertainty.
- Reference objects may have none, one or more ordinal or numerical characteristics, whether continuous or discrete (furnace temperature, radar echo azimuth, number of wheels of a vehicle). Likewise, reference objects may have none, one or more nominal characteristics, such as name, gender, or class of membership. These additional features can be visualized on the map by a specific appearance, by example in terms of size, shape, texture, color, or as a region associated with this reference object.
- the invention allows the user to visually and comprehensively grasp the proximity of the request object to the reference objects in terms of similarity. Thus, it helps him to make a decision as to the nature of this object and the treatment that may suit him.
- the association of a cartographic visualization whose position of the reference objects is stable, a similarity measure to be displayed on the map, as well as the absence of positioning of the query object on the map make the invention more particularly exploitable in the field of decision support in discrimination and in the field of anomaly detection of a request object relative to reference objects.
- the main advantage of the invention is that it presents a map of the reference objects such that the position of these objects or zones used to represent them is fixed and independent of the request object.
- This map is therefore a stable basis for visual apprehension of the universe of reference objects, as well as easy memorization of this representation. This stability allows the user to focus on the similarities between the request object and the reference objects rather than between the reference objects themselves, since it is not disturbed by object position changes. reference.
- the representation of the similarity between the request object and the reference objects by a visual parameter of these reference objects, other than their position offers an immediate visual perception of the most similar or the most different reference objects of the reference objects. the request object.
- a system implementing the method according to the invention described above can be implemented on most computers equipped with a graphic display device.
- the main advantage of the invention described above is that no decision is made: there is no confidence index, no combination of logical rules, nor any probability of global belonging to classes, all information which the user does not control the provenance and interpretation. It is the similarity measures between the request object and each reference object that are displayed. Above all, it is a measure of similarity between the request object and each reference object that is displayed, and not just a characteristic of the reference objects independent of the request object. This point is particularly advantageous when the objects have more than one characteristic, making it difficult to visualize on the same map these multiple characteristics for each object and complicating the visual comparison with the characteristics of the request object.
- the invention makes it possible to visualize, without deformation, the raw similarity measurements provided as input.
- the measure will be known to the user or at least it will be intelligible: there is no bias due to another treatment not controlled by the user. This renders intelligible the information displayed and is conducive to the user having confidence in this information.
- the invention also applies to objects that do not necessarily have a natural representation in the form of a card, because the representation of the resemblance between the request object and the reference objects does not depend on this positioning.
- the invention can therefore not only be applied to objects positioned on the card by any automatic or manual means, but it can also be applied to objects whose graphic representation in the form of a card is predefined, such as the borders of the card.
- each zone corresponds to a reference object.
- the displayed measurement is a measure of similarity between the request object and the reference objects, which makes it possible to position the first mentally relative to the second, whereas the cards according to the prior art represent information specific to the reference objects represented. regardless of the request object.
- the fields of application of the invention are vast, the method according to the invention being generic and can therefore be applied to any field involving a decision support system in discrimination, including shape recognition systems.
- the invention is applicable in the field of medical diagnostic aid, such as the diagnosis of melanoma.
- diagnosis of melanoma is very difficult for general practitioners.
- Decision support tools can assist general practitioners in their choice to send or not to consult a dermatologist.
- the "query" melanoma stains reference melanomas and allows the doctor to determine its severity.
- the invention is applicable in the field of the search for the origin of seismic events, such as the determination of their natural or anthropogenic origin (e.g. career shots). It is a routine work done by geophysical analysts from signals captured on multiple measurement stations. The analyst visualizes a map of the events usually encountered, grouped spatially according to their origin. The event being analyzed colors similar reference events on the map, thus helping the analyst to determine the origin of the event.
- origin of seismic events such as the determination of their natural or anthropogenic origin (e.g. career shots).
- the invention is applicable in the field of marketing, such as customer behavior analysis.
- Reference customers can be viewed and grouped by category on a map, with each category corresponding to a particular target to whom specific advertising messages are sent.
- a new customer is visualized according to their proximity to the reference customers, which makes it possible to recognize the category or categories of which it is the closest.
- the invention is applicable in the field of risk assessment in credit, stock market or insurance. It is a question of assessing the risks of drifting of the financial situation of a customer to define the type of credit or the rate of risk to be applied to him.
- the invention is applicable in the field of biometrics.
- An individual can be identified by a photograph of his face or a fingerprint. These elements can be compared to those of reference positioned on a map. The interviewer analyst quickly sees if the individual is similar to one or more reference individuals or conversely completely new.
- the invention is applicable in the field of industrial or computer security.
- the operator responsible for monitoring the operation of the plant displays a map of the different reference states usually measured during normal operation.
- the current state of operation is displayed as a coloring of the reference states related to their similarity to this current state. If the current state appears to move further and further away from the reference states, the operator sees it and initiates the appropriate shutdown, evacuation or simple control procedures.
- an intruder can be detected in a computer system whose behavior does not resemble normal referenced behavior.
- the invention is applicable in the field of transport, logistics or predictive maintenance. It is then a matter of monitoring the state of the flows and visual detection of drifts compared to a reference situation.
- the invention is applicable in the field of classification of digital documents, Internet favorites, web pages or personal folders.
- a user who visualizes a new website and wishes to add it to the list of his favorite sites is then presented sites already present in this list in the form of a map.
- the new site then colors the preferred sites according to their similarity with it, which allows the user to decide on the most appropriate category or categories to classify it, or the creation of a new category.
- the invention is applicable in the field of consumer assistance in the choice of a complex product defined by multiple features, such as a TV, a washing machine, a mobile phone, a computer, a car, a house, an insurance, an investment product, a mobile phone subscription.
- typical package offers (references) are represented on a card, and the customer is asked to define his type of consumption (so his ideal package).
- the invention then makes it possible to present to the customer the packages closest to his ideal package, the organization in card to clearly distinguish packages close to the ideal customer over others.
- the invention then also makes it possible to distinguish the different families of packages that differ drastically according to characteristics that the customer would not have informed (price, internet option ). This allows the customer to focus on each of these families of offers very quickly and visualize "where" it is in the maquis information through the map.
- thumbnails are for illustrative purposes only. Indeed, the present invention is also applicable to all kinds of data, including digitized data. These digitized data may include physical characteristics measurements taken on a wide variety of objects other than photos, be they physical objects, individuals, system states, or even a group. such objects, individuals or states, whose physical characteristics are measured.
- these digitized data may include scalars, i.e., real numbers, such as measurements provided by a sensor.
- these digitized data can also include symbols (element of an alphabet) as an elementary value of a finite set (letter of a word, name of an object, etc.).
- These digitized data can also include vectors, such as a measurement of a sensor accompanied by its uncertainty or a set of measurements from a sensor network or a signal (sequence of measurements, flows, etc.) or a set of values from a database or a word, a sentence, a text or a set of standardized measures (proportions) or any set of scalar or symbolic data.
- vectors such as a measurement of a sensor accompanied by its uncertainty or a set of measurements from a sensor network or a signal (sequence of measurements, flows, etc.) or a set of values from a database or a word, a sentence, a text or a set of standardized measures (proportions) or any set of scalar or symbolic data.
- These digitized data can also include matrices, such as a black-and-white flat image or a set of signals from a sensor network or genetic data or any set of vector data.
- This digitized data may also include multi-dimensional arrays, such as a sequence of images (video) or a multi-spectral image (satellite image) or a color image (photograph, simulation result) or a 3D image (scanner) or a multi-dimensional mesh (simulation model) or any set of matrix data or smaller multi-dimensional arrays.
- multi-dimensional arrays such as a sequence of images (video) or a multi-spectral image (satellite image) or a color image (photograph, simulation result) or a 3D image (scanner) or a multi-dimensional mesh (simulation model) or any set of matrix data or smaller multi-dimensional arrays.
- This digitized data can also include graphs and networks, such as a social network or the Internet or a transport network (road traffic, information, energy, etc.) or a network of interactions (proteins, genes) or a network of sensors or a numerical modeling mesh (modeling 2D, 3D, 3D with the time, etc).
- graphs and networks such as a social network or the Internet or a transport network (road traffic, information, energy, etc.) or a network of interactions (proteins, genes) or a network of sensors or a numerical modeling mesh (modeling 2D, 3D, 3D with the time, etc).
- digitized data may also include cellular complexes or hypergraphs, such as a digital modeling mesh (virtual objects, multi-physics modeling, animation films) or biological or molecular or physical or climate or mechanical or chemical models.
- a digital modeling mesh virtual objects, multi-physics modeling, animation films
- biological or molecular or physical or climate or mechanical or chemical models such as a digital modeling mesh (virtual objects, multi-physics modeling, animation films) or biological or molecular or physical or climate or mechanical or chemical models.
- This digitized data can also include complex data such as multimedia documents (organized set of texts, videos, audio signals, etc.) or a collection of documents or any set of organized documents (library).
- multimedia documents organized set of texts, videos, audio signals, etc.
- library any set of organized documents
- This digitized data may also include service subscription contracts, such as telephone subscription contracts, for example.
- service subscription contracts such as telephone subscription contracts, for example.
- the method and the system according to the present invention could then advantageously make it possible to choose the most suitable telephone package according to the profile of the user.
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- Data Mining & Analysis (AREA)
- General Physics & Mathematics (AREA)
- Databases & Information Systems (AREA)
- General Engineering & Computer Science (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Mathematical Physics (AREA)
- Pure & Applied Mathematics (AREA)
- Computational Mathematics (AREA)
- Mathematical Optimization (AREA)
- Mathematical Analysis (AREA)
- Library & Information Science (AREA)
- Life Sciences & Earth Sciences (AREA)
- Bioinformatics & Computational Biology (AREA)
- Evolutionary Biology (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Artificial Intelligence (AREA)
- General Business, Economics & Management (AREA)
- Business, Economics & Management (AREA)
- Evolutionary Computation (AREA)
- Software Systems (AREA)
- Computing Systems (AREA)
- Algebra (AREA)
- Image Analysis (AREA)
- Information Retrieval, Db Structures And Fs Structures Therefor (AREA)
Abstract
Description
Claims
Priority Applications (4)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
US13/503,621 US9576223B2 (en) | 2009-10-23 | 2010-10-22 | Method and system for evaluating the resemblance of a query object to reference objects |
JP2012534714A JP2013508840A (ja) | 2009-10-23 | 2010-10-22 | 参照オブジェクトに対するクエリオブジェクトの類似性を評価するための方法およびシステム |
CA2778651A CA2778651C (fr) | 2009-10-23 | 2010-10-22 | Methode et systeme pour evaluer la ressemblance d'un objet requete a des objets de reference |
EP10768037A EP2491518A2 (fr) | 2009-10-23 | 2010-10-22 | Methode et système pour évaluer la ressemblance d'un objet requête à des objets de référence |
Applications Claiming Priority (2)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
FR0905117A FR2951839B1 (fr) | 2009-10-23 | 2009-10-23 | Methode pour evaluer la ressemblance d'un objet requete a des objets de reference |
FR0905117 | 2009-10-23 |
Publications (2)
Publication Number | Publication Date |
---|---|
WO2011048219A2 true WO2011048219A2 (fr) | 2011-04-28 |
WO2011048219A3 WO2011048219A3 (fr) | 2011-06-16 |
Family
ID=42321033
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
PCT/EP2010/065997 WO2011048219A2 (fr) | 2009-10-23 | 2010-10-22 | Methode et systeme pour evaluer la ressemblance d'un objet requete a des objets de reference |
Country Status (6)
Country | Link |
---|---|
US (1) | US9576223B2 (fr) |
EP (1) | EP2491518A2 (fr) |
JP (1) | JP2013508840A (fr) |
CA (1) | CA2778651C (fr) |
FR (1) | FR2951839B1 (fr) |
WO (1) | WO2011048219A2 (fr) |
Cited By (1)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
FR3060794A1 (fr) * | 2016-12-21 | 2018-06-22 | Commissariat A L'energie Atomique Et Aux Energies Alternatives | Procede de determination de l'etat d'un systeme, procede de determination d'une methode de projection optimale et dispositif mettant en œuvre lesdits procedes |
Families Citing this family (20)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US10771247B2 (en) | 2013-03-15 | 2020-09-08 | Commerce Signals, Inc. | Key pair platform and system to manage federated trust networks in distributed advertising |
US10803512B2 (en) * | 2013-03-15 | 2020-10-13 | Commerce Signals, Inc. | Graphical user interface for object discovery and mapping in open systems |
US11222346B2 (en) | 2013-03-15 | 2022-01-11 | Commerce Signals, Inc. | Method and systems for distributed signals for use with advertising |
US20140378810A1 (en) | 2013-04-18 | 2014-12-25 | Digimarc Corporation | Physiologic data acquisition and analysis |
US10425702B2 (en) | 2015-09-30 | 2019-09-24 | Sensormatic Electronics, LLC | Sensor packs that are configured based on business application |
US11436911B2 (en) | 2015-09-30 | 2022-09-06 | Johnson Controls Tyco IP Holdings LLP | Sensor based system and method for premises safety and operational profiling based on drift analysis |
US11151654B2 (en) | 2015-09-30 | 2021-10-19 | Johnson Controls Tyco IP Holdings LLP | System and method for determining risk profile, adjusting insurance premiums and automatically collecting premiums based on sensor data |
US10902524B2 (en) * | 2015-09-30 | 2021-01-26 | Sensormatic Electronics, LLC | Sensor based system and method for augmenting underwriting of insurance policies |
US10552914B2 (en) | 2016-05-05 | 2020-02-04 | Sensormatic Electronics, LLC | Method and apparatus for evaluating risk based on sensor monitoring |
US10810676B2 (en) | 2016-06-06 | 2020-10-20 | Sensormatic Electronics, LLC | Method and apparatus for increasing the density of data surrounding an event |
US10740647B2 (en) | 2018-03-14 | 2020-08-11 | Adobe Inc. | Detecting objects using a weakly supervised model |
US11107219B2 (en) | 2019-07-22 | 2021-08-31 | Adobe Inc. | Utilizing object attribute detection models to automatically select instances of detected objects in images |
US11631234B2 (en) | 2019-07-22 | 2023-04-18 | Adobe, Inc. | Automatically detecting user-requested objects in images |
US11302033B2 (en) | 2019-07-22 | 2022-04-12 | Adobe Inc. | Classifying colors of objects in digital images |
US11468550B2 (en) | 2019-07-22 | 2022-10-11 | Adobe Inc. | Utilizing object attribute detection models to automatically select instances of detected objects in images |
US11468110B2 (en) | 2020-02-25 | 2022-10-11 | Adobe Inc. | Utilizing natural language processing and multiple object detection models to automatically select objects in images |
US11055566B1 (en) * | 2020-03-12 | 2021-07-06 | Adobe Inc. | Utilizing a large-scale object detector to automatically select objects in digital images |
US11587234B2 (en) | 2021-01-15 | 2023-02-21 | Adobe Inc. | Generating class-agnostic object masks in digital images |
US11972569B2 (en) | 2021-01-26 | 2024-04-30 | Adobe Inc. | Segmenting objects in digital images utilizing a multi-object segmentation model framework |
US20230376164A1 (en) * | 2022-05-18 | 2023-11-23 | Qatar Foundation For Education, Science And Community Development | Active learning for scalable arrangement and grouping of data |
Family Cites Families (10)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US6539115B2 (en) * | 1997-02-12 | 2003-03-25 | Fujitsu Limited | Pattern recognition device for performing classification using a candidate table and method thereof |
US7272593B1 (en) * | 1999-01-26 | 2007-09-18 | International Business Machines Corporation | Method and apparatus for similarity retrieval from iterative refinement |
US6446060B1 (en) * | 1999-01-26 | 2002-09-03 | International Business Machines Corporation | System and method for sequential processing for content-based retrieval of composite objects |
US20090006156A1 (en) * | 2007-01-26 | 2009-01-01 | Herbert Dennis Hunt | Associating a granting matrix with an analytic platform |
US7716226B2 (en) * | 2005-09-27 | 2010-05-11 | Patentratings, Llc | Method and system for probabilistically quantifying and visualizing relevance between two or more citationally or contextually related data objects |
US8005643B2 (en) * | 2007-06-26 | 2011-08-23 | Endeca Technologies, Inc. | System and method for measuring the quality of document sets |
WO2009143040A1 (fr) * | 2008-05-19 | 2009-11-26 | Disetronic Medical Systems Ag | Outil de recherche informatique pour l'organisation, la visualisation et l'analyse de données cliniques liées au métabolisme et son procédé |
US8559671B2 (en) * | 2008-12-18 | 2013-10-15 | The Regents Of The University Of California | Training-free generic object detection in 2-D and 3-D using locally adaptive regression kernels |
US8948515B2 (en) * | 2010-03-08 | 2015-02-03 | Sightera Technologies Ltd. | Method and system for classifying one or more images |
US8498949B2 (en) * | 2010-08-11 | 2013-07-30 | Seiko Epson Corporation | Supervised nonnegative matrix factorization |
-
2009
- 2009-10-23 FR FR0905117A patent/FR2951839B1/fr active Active
-
2010
- 2010-10-22 EP EP10768037A patent/EP2491518A2/fr not_active Ceased
- 2010-10-22 CA CA2778651A patent/CA2778651C/fr active Active
- 2010-10-22 US US13/503,621 patent/US9576223B2/en active Active
- 2010-10-22 WO PCT/EP2010/065997 patent/WO2011048219A2/fr active Application Filing
- 2010-10-22 JP JP2012534714A patent/JP2013508840A/ja active Pending
Non-Patent Citations (1)
Title |
---|
AUPETIT M.: "Visualizing distortions and recovering topology in continuous projection techniques", NEUROCOMPUTING, vol. 10, no. 7-9, 2007, pages 1304 - 1330, XP022067245, DOI: doi:10.1016/j.neucom.2006.11.018 |
Cited By (2)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
FR3060794A1 (fr) * | 2016-12-21 | 2018-06-22 | Commissariat A L'energie Atomique Et Aux Energies Alternatives | Procede de determination de l'etat d'un systeme, procede de determination d'une methode de projection optimale et dispositif mettant en œuvre lesdits procedes |
EP3340065A1 (fr) * | 2016-12-21 | 2018-06-27 | Commissariat à l'Energie Atomique et aux Energies Alternatives | Procédé de détermination de l'état d'un système, procédé de détermination d'une méthode de projection optimale et dispositif mettant en oeuvre lesdits procédés |
Also Published As
Publication number | Publication date |
---|---|
CA2778651A1 (fr) | 2011-04-28 |
JP2013508840A (ja) | 2013-03-07 |
WO2011048219A3 (fr) | 2011-06-16 |
CA2778651C (fr) | 2019-01-29 |
FR2951839A1 (fr) | 2011-04-29 |
EP2491518A2 (fr) | 2012-08-29 |
FR2951839B1 (fr) | 2021-06-11 |
US20130066592A1 (en) | 2013-03-14 |
US9576223B2 (en) | 2017-02-21 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
WO2011048219A2 (fr) | Methode et systeme pour evaluer la ressemblance d'un objet requete a des objets de reference | |
Patra et al. | Histogram thresholding for unsupervised change detection of remote sensing images | |
Pastore et al. | Annotation-free learning of plankton for classification and anomaly detection | |
Ke et al. | A comparison of three methods for automatic tree crown detection and delineation from high spatial resolution imagery | |
Khoshboresh-Masouleh et al. | Multiscale building segmentation based on deep learning for remote sensing RGB images from different sensors | |
Hamedianfar et al. | Development of fuzzy rule-based parameters for urban object-oriented classification using very high resolution imagery | |
FR2991078A1 (fr) | Selection d'image basee sur un style photographique | |
Ruz et al. | Automated visual inspection system for wood defect classification using computational intelligence techniques | |
US7606779B2 (en) | Methods and system for data aggregation of physical samples | |
Chatterjee et al. | Intelligent Road Maintenance: a Machine Learning Approach for surface Defect Detection. | |
CN113313149B (zh) | 一种基于注意力机制和度量学习的菜品识别方法 | |
Su et al. | Classification of MODIS images combining surface temperature and texture features using the Support Vector Machine method for estimation of the extent of sea ice in the frozen Bohai Bay, China | |
Zhang et al. | Feature extraction for high-resolution imagery based on human visual perception | |
Sangeetha et al. | Detection of pest and disease in banana leaf using convolution Random Forest | |
Gardella et al. | Forgery detection in digital images by multi-scale noise estimation | |
Paul et al. | Canopy averaged chlorophyll content prediction of pear trees using convolutional autoencoder on hyperspectral data | |
Alshehri | A content-based image retrieval method using neural network-based prediction technique | |
Rehman et al. | Long short term memory deep net performance on fused Planet-Scope and Sentinel-2 imagery for detection of agricultural crop | |
Kanaev et al. | Object level HSI-LIDAR data fusion for automated detection of difficult targets | |
EP2491517B1 (fr) | Méthode et système pour évaluer la classe d'une donnée de test dans un espace de données de grande dimension | |
Ghate et al. | Forest wildfire detection and forecasting utilizing machine learning and image processing | |
Abishek et al. | Soil Texture Prediction Using Machine Learning Approach for Sustainable Soil Health Management | |
Abraham | Digital image forgery detection approaches: A review and analysis | |
Vudattu et al. | Change Detection in Synthetic Aperture Radar Images based on a Spatial Pyramid Pooling Attention Network (SPPANet) | |
Naguib et al. | AI Assistance for Home Surveillance |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
121 | Ep: the epo has been informed by wipo that ep was designated in this application |
Ref document number: 10768037 Country of ref document: EP Kind code of ref document: A1 |
|
WWE | Wipo information: entry into national phase |
Ref document number: 2012534714 Country of ref document: JP |
|
NENP | Non-entry into the national phase |
Ref country code: DE |
|
WWE | Wipo information: entry into national phase |
Ref document number: 2778651 Country of ref document: CA |
|
WWE | Wipo information: entry into national phase |
Ref document number: 3611/DELNP/2012 Country of ref document: IN |
|
WWE | Wipo information: entry into national phase |
Ref document number: 2010768037 Country of ref document: EP |
|
WWE | Wipo information: entry into national phase |
Ref document number: 13503621 Country of ref document: US |