CN113743266B - Human face recognition method based on artificial myxobacteria - Google Patents
Human face recognition method based on artificial myxobacteria Download PDFInfo
- Publication number
- CN113743266B CN113743266B CN202110977206.XA CN202110977206A CN113743266B CN 113743266 B CN113743266 B CN 113743266B CN 202110977206 A CN202110977206 A CN 202110977206A CN 113743266 B CN113743266 B CN 113743266B
- Authority
- CN
- China
- Prior art keywords
- face
- artificial
- feature
- matching
- human
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Active
Links
- 241000282414 Homo sapiens Species 0.000 title claims abstract description 119
- 238000000034 method Methods 0.000 title claims abstract description 66
- 241000863434 Myxococcales Species 0.000 title claims description 35
- 239000013598 vector Substances 0.000 claims abstract description 142
- 230000006870 function Effects 0.000 claims abstract description 34
- 238000004422 calculation algorithm Methods 0.000 claims abstract description 26
- 230000008569 process Effects 0.000 claims abstract description 25
- 230000008602 contraction Effects 0.000 claims abstract description 17
- 230000019637 foraging behavior Effects 0.000 claims abstract description 12
- 230000001815 facial effect Effects 0.000 claims description 49
- 235000016709 nutrition Nutrition 0.000 claims description 42
- 230000035764 nutrition Effects 0.000 claims description 37
- 238000004364 calculation method Methods 0.000 claims description 26
- 235000015097 nutrients Nutrition 0.000 claims description 23
- 238000012549 training Methods 0.000 claims description 18
- 230000002431 foraging effect Effects 0.000 claims description 16
- 239000011159 matrix material Substances 0.000 claims description 14
- 238000007781 pre-processing Methods 0.000 claims description 12
- 210000000056 organ Anatomy 0.000 claims description 10
- 108010078777 Colistin Proteins 0.000 claims description 9
- 229960003346 colistin Drugs 0.000 claims description 9
- 230000014509 gene expression Effects 0.000 claims description 9
- JORAUNFTUVJTNG-BSTBCYLQSA-N n-[(2s)-4-amino-1-[[(2s,3r)-1-[[(2s)-4-amino-1-oxo-1-[[(3s,6s,9s,12s,15r,18s,21s)-6,9,18-tris(2-aminoethyl)-3-[(1r)-1-hydroxyethyl]-12,15-bis(2-methylpropyl)-2,5,8,11,14,17,20-heptaoxo-1,4,7,10,13,16,19-heptazacyclotricos-21-yl]amino]butan-2-yl]amino]-3-h Chemical compound CC(C)CCCCC(=O)N[C@@H](CCN)C(=O)N[C@H]([C@@H](C)O)CN[C@@H](CCN)C(=O)N[C@H]1CCNC(=O)[C@H]([C@@H](C)O)NC(=O)[C@H](CCN)NC(=O)[C@H](CCN)NC(=O)[C@H](CC(C)C)NC(=O)[C@@H](CC(C)C)NC(=O)[C@H](CCN)NC1=O.CCC(C)CCCCC(=O)N[C@@H](CCN)C(=O)N[C@H]([C@@H](C)O)CN[C@@H](CCN)C(=O)N[C@H]1CCNC(=O)[C@H]([C@@H](C)O)NC(=O)[C@H](CCN)NC(=O)[C@H](CCN)NC(=O)[C@H](CC(C)C)NC(=O)[C@@H](CC(C)C)NC(=O)[C@H](CCN)NC1=O JORAUNFTUVJTNG-BSTBCYLQSA-N 0.000 claims description 9
- XDJYMJULXQKGMM-UHFFFAOYSA-N polymyxin E1 Natural products CCC(C)CCCCC(=O)NC(CCN)C(=O)NC(C(C)O)C(=O)NC(CCN)C(=O)NC1CCNC(=O)C(C(C)O)NC(=O)C(CCN)NC(=O)C(CCN)NC(=O)C(CC(C)C)NC(=O)C(CC(C)C)NC(=O)C(CCN)NC1=O XDJYMJULXQKGMM-UHFFFAOYSA-N 0.000 claims description 9
- KNIWPHSUTGNZST-UHFFFAOYSA-N polymyxin E2 Natural products CC(C)CCCCC(=O)NC(CCN)C(=O)NC(C(C)O)C(=O)NC(CCN)C(=O)NC1CCNC(=O)C(C(C)O)NC(=O)C(CCN)NC(=O)C(CCN)NC(=O)C(CC(C)C)NC(=O)C(CC(C)C)NC(=O)C(CCN)NC1=O KNIWPHSUTGNZST-UHFFFAOYSA-N 0.000 claims description 9
- 210000001938 protoplast Anatomy 0.000 claims description 6
- 230000009466 transformation Effects 0.000 claims description 6
- 238000012937 correction Methods 0.000 claims description 5
- 238000000605 extraction Methods 0.000 claims description 5
- 238000010606 normalization Methods 0.000 claims description 5
- 210000003855 cell nucleus Anatomy 0.000 claims description 4
- 210000000805 cytoplasm Anatomy 0.000 claims description 4
- 238000001914 filtration Methods 0.000 claims description 4
- 238000013507 mapping Methods 0.000 claims description 4
- 238000004088 simulation Methods 0.000 claims description 4
- 241000235395 Mucor Species 0.000 claims description 3
- 230000003232 mucoadhesive effect Effects 0.000 claims description 3
- 210000000697 sensory organ Anatomy 0.000 claims description 3
- 201000003883 Cystic fibrosis Diseases 0.000 claims description 2
- 230000009471 action Effects 0.000 claims description 2
- 230000006399 behavior Effects 0.000 claims description 2
- 238000012512 characterization method Methods 0.000 claims description 2
- 210000001508 eye Anatomy 0.000 claims description 2
- 230000008713 feedback mechanism Effects 0.000 claims description 2
- BTCSSZJGUNDROE-UHFFFAOYSA-N gamma-aminobutyric acid Chemical compound NCCCC(O)=O BTCSSZJGUNDROE-UHFFFAOYSA-N 0.000 claims description 2
- 230000003068 static effect Effects 0.000 claims description 2
- 230000004634 feeding behavior Effects 0.000 claims 1
- 238000005457 optimization Methods 0.000 abstract description 3
- 238000012360 testing method Methods 0.000 description 5
- 238000005516 engineering process Methods 0.000 description 4
- 230000037406 food intake Effects 0.000 description 4
- 239000004459 forage Substances 0.000 description 4
- 206010028116 Mucosal inflammation Diseases 0.000 description 3
- 201000010927 Mucositis Diseases 0.000 description 3
- 238000011160 research Methods 0.000 description 3
- 241000894006 Bacteria Species 0.000 description 2
- 238000013135 deep learning Methods 0.000 description 2
- 230000000694 effects Effects 0.000 description 2
- 230000007613 environmental effect Effects 0.000 description 2
- 239000000284 extract Substances 0.000 description 2
- 238000001727 in vivo Methods 0.000 description 2
- 238000005259 measurement Methods 0.000 description 2
- 230000007246 mechanism Effects 0.000 description 2
- 239000011664 nicotinic acid Substances 0.000 description 2
- 230000009024 positive feedback mechanism Effects 0.000 description 2
- 238000012545 processing Methods 0.000 description 2
- 238000005070 sampling Methods 0.000 description 2
- 230000035945 sensitivity Effects 0.000 description 2
- 238000012935 Averaging Methods 0.000 description 1
- 238000010521 absorption reaction Methods 0.000 description 1
- 230000008859 change Effects 0.000 description 1
- 238000013527 convolutional neural network Methods 0.000 description 1
- 238000013075 data extraction Methods 0.000 description 1
- 230000007812 deficiency Effects 0.000 description 1
- 238000011161 development Methods 0.000 description 1
- 230000029087 digestion Effects 0.000 description 1
- 210000002569 neuron Anatomy 0.000 description 1
- 238000011002 quantification Methods 0.000 description 1
- 238000010845 search algorithm Methods 0.000 description 1
- 230000003313 weakening effect Effects 0.000 description 1
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
- G06F18/214—Generating training patterns; Bootstrap methods, e.g. bagging or boosting
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/22—Matching criteria, e.g. proximity measures
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/045—Combinations of networks
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- Data Mining & Analysis (AREA)
- Life Sciences & Earth Sciences (AREA)
- Artificial Intelligence (AREA)
- General Physics & Mathematics (AREA)
- General Engineering & Computer Science (AREA)
- Evolutionary Computation (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Computational Linguistics (AREA)
- Bioinformatics & Computational Biology (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Health & Medical Sciences (AREA)
- Biomedical Technology (AREA)
- Biophysics (AREA)
- Evolutionary Biology (AREA)
- General Health & Medical Sciences (AREA)
- Molecular Biology (AREA)
- Computing Systems (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- Image Analysis (AREA)
- Image Processing (AREA)
Abstract
The human face recognition method based on the artificial mucosae is used for simulating mucosae foraging behaviors to match and recognize human face images, and extracting human face feature vectors through the expansion and contraction processes of the artificial mucosae in the human face images; the specific algorithm comprises the following steps: step 1, initializing a face image and artificial mucobacteria; step 2, extracting features of the face image by using artificial mucobacteria; and 3, matching and identifying the face image by using the artificial mucosae. Expanding the mucomorph of the artificial mucobacteria around the face image, searching the face characteristic pixel points, and adding the searched face characteristic pixel points into a face characteristic pixel point set of the mucomorph; then, the artificial mucobacteria start to continuously shrink, namely, the external food sources or the human face characteristic pixel points are digested, so that the human face matching and recognition functions are realized. The method can efficiently finish the optimization solution of face recognition.
Description
Technical Field
The invention relates to the technical field of face recognition, in particular to a face recognition method based on artificial mucosae.
Background
Face recognition technology has been an attractive technology in the field of computer vision, and a series of sub-algorithms generated around the technology in recent years also grow endlessly, such as face recognition algorithms based on convolutional neural networks, face recognition algorithms based on deep learning, and the like.
Although the conventional face recognition algorithm can play a role in recognition, the following problems still exist: firstly, the extraction algorithm of the characteristic value of the face image used by the traditional face recognition algorithm is rapidly increased along with the increase of computing nodes, so that the method is difficult to adapt to the rapidly increased picture requirement, and deeper depth characteristics with semantics are inconvenient to obtain from the original image. Secondly, in order to obtain a better recognition effect, the traditional face recognition algorithm is required to be combined with artificial feature assistance, but in the process, the artificial feature often brings interference and errors which do not conform to expected human factors, and the accuracy of face recognition is reduced. Thirdly, under the condition of reducing artificial interference, the traditional face recognition surface often presents difficulty and deficiency of low algorithm performance for a larger data set. For example, the deep learning algorithm is used for face recognition, and because of the large number of neurons, the operation time is relatively long, and a plurality of pieces of face data are needed for training.
With the development of biotechnology, human beings have been increasingly intensively studied for biology with simulation functions. Slime is a type of organism with special biological simulation properties. The scientific research shows that the myxobacteria can carry out foraging activity through the contraction of protoplasts and have high sensitivity to food in the foraging activity. The characteristics exhibited by japanese scientists in the foraging of myxobacteria in 2000 have been used to simulate the best performing tokyo rail network. Scientific researches find that the distributed search characteristics with higher performance than that of a computer exist in foraging behaviors of the brain-free biological mucosae.
Disclosure of Invention
The invention provides a human face recognition method based on artificial mucosae, which is based on mathematics and biological models, and constructs biological and mathematical theories based on the sensitivity mechanism of mucosae to food by utilizing the rule in the mucosae foraging process, and constructs a human face recognition algorithm and model based on the biological and mathematical theories. Compared with the face recognition algorithm widely applied at present, the mucoadhesive bionic algorithm has the advantages of high speed, low cost, good robustness and the like in the face recognition process.
The technical scheme adopted by the invention is as follows:
the human face recognition method based on the artificial mucosae simulates mucosae foraging behaviors to match and recognize the human face images, and extracts human face feature vectors through the expansion and contraction processes of the artificial mucosae in the human face images; simulating the initial position of a single mucor as a center pixel point which is the center of a face image, and simulating the face characteristic pixel point as an external food source; comparing and matching the extracted face feature vector with a face feature model stored in a face feature database, wherein the size of the matching value can be simulated into the size of an external food source, namely the larger the face matching value is, the larger the food source is, otherwise, the smaller the face matching value is, the smaller the food source is; simulating a virtual edge between two pixel points in the face feature vector as a deformation body, and simulating a pixel point difference value on the deformation body as a nutrition element value; the expansion behavior of the deformable body is simulated as searching the face feature vector, and the contraction of the deformable body is simulated as matching the face feature vector.
The human face recognition method based on the artificial mucosae simulates mucosae foraging behaviors to match and recognize the human face images, and extracts human face feature vectors through the expansion and contraction processes of the artificial mucosae in the human face images; the method comprises the following steps:
step 1, initializing a face image and artificial mucobacteria;
step 2, extracting features of the face image by using artificial mucobacteria;
and 3, matching and identifying the face image by using the artificial mucosae. Expanding the mucomorph of the artificial mucobacteria around the face image, searching the face characteristic pixel points, and adding the searched face characteristic pixel points into a face characteristic pixel point set of the mucomorph; then, the artificial mucobacteria start to continuously shrink, namely, the external food sources or the human face characteristic pixel points are digested, so that the human face matching and recognition functions are realized. The method comprises the following steps:
step 1, initializing a face image and artificial mucobacteria;
collecting a face image, preprocessing the face image, performing light compensation, gray level transformation, histogram equalization, normalization, geometric correction, filtering and sharpening on each pixel point of the face image, and constructing a face image preprocessing vector according to a preprocessing result; initializing the artificial mucobacteria, fixing the cell nucleus of the artificial mucobacteria at the center point of the face image, and initializing the quantity of the mucoshape bodies of the artificial mucobacteria by using a random number; setting a human face characteristic pixel point set in the deformable body as an empty set, wherein human face characteristic vectors formed by all human face characteristic pixel point sets are empty sets, namely the deformable body does not find human face characteristic pixel points or external food sources during initialization; removing the values of nutrient elements in the visco-deformation body, wherein the transport of the nutrient elements is not carried out in the visco-deformation body temporarily; after the initialization is completed, entering a step 2;
Step 2, extracting characteristics of the human face image by using artificial mucobacteria:
expanding the mucomorph body of the artificial mucobacteria on the face image everywhere, searching face feature pixel points, adding the searched face feature pixel points into a face feature pixel point set of the mucomorph body, updating the face feature pixel point set of all the mucomorph bodies, and updating face feature vectors; the artificial mucobacteria are used for extracting the facial features, the mucomorph body can be connected with different facial feature pixel points for transporting nutrition elements, the larger external food source or the facial feature pixel points with higher matching values can provide more nutrition elements, and the smaller the external food source or the facial feature pixel points with lower matching values provide fewer nutrition elements, the nutrition element values in the mucomorph body are updated; the mucomorph body is continuously expanded until all face feature pixel points are found, the artificial mucobacteria expansion process is finished, the face feature extraction is finished, and the step 3 is entered;
step 3, matching and identifying the face image by using artificial mucobacteria:
in the step, the artificial mucobacteria start to continuously shrink, namely, the external food sources or the human face characteristic pixel points are digested, so that the human face matching and recognition functions are realized; the mucomorph starts to absorb the nutrition elements on the external food source or the human face characteristic pixel points, and the simulation real mucobacteria transport the nutrition elements into the human body, so that the matching values of the nutrition elements on the external food source or the human face characteristic pixel points start to be reduced; the face characteristic pixel points with lower matching values can be absorbed by the nutrition elements or the matching values to disappear, and the face characteristic pixel points with larger matching values can provide the nutrition elements or the matching values to strengthen; updating the nutrition element values in the sticky deformation body and updating the matching values on the face characteristic pixel points; the mucosae body continuously contracts until the matching values on all the face characteristic pixel points accord with a preset threshold value, the artificial mucobacteria contraction process is finished, the face matching and recognition process is finished, and the step 4 is entered;
Step 4, outputting the human face recognition result of the artificial colistin:
outputting face images after face recognition, identifying faces in the images, outputting a matched and recognized face characteristic pixel point matrix, and outputting related calculation parameters of artificial mucobacteria, wherein the related calculation parameters comprise nutrition element values of mucomorphs, and matching values of all face characteristic pixel points.
The step 1, the initialization of the face image and the artificial mucosae, comprises three substeps:
substep 1-1: collecting a face image; different face images can be acquired by using various shooting devices, including a digital camera for acquiring a static face image, a camera for acquiring a dynamic face image, a camera array for acquiring face images of different positions and angles, and a camera with different focuses for simultaneously acquiring different expressions of the face; when a user enters the acquisition range of the shooting equipment, the shooting equipment can automatically search and acquire face images of the user; after finishing face image acquisition in the substep 1-1, turning to the substep 1-2 for pretreatment;
substep 1-2: preprocessing a face image; the method mainly comprises the steps of light compensation, gray level transformation, histogram equalization, image normalization, geometric correction, image filtering and sharpening of a face image; different preprocessing functions are adopted to eliminate possible deviation and noise in the face image acquired in the substep 1-1, so that the quality of the face image is improved; initializing a face feature database, and initializing a face feature matching function for face feature matching and matching value calculation;
Substep 1-3: initializing artificial mucosae parameters; setting cell nuclei of the artificial mucobacteria on a central pixel point of a face image, setting the quantity of mucoshape bodies of the artificial mucobacteria, setting learning factors of the artificial mucobacteria, eliminating nutrition element values in the mucoshape bodies to be 0, eliminating face characteristic pixel point sets in the mucoshape bodies to be empty sets, and eliminating face characteristic vectors to be 0; and importing a face feature database and a face feature model.
Step 2, the characteristic extraction of the human face image by the artificial mucosae comprises three substeps:
substep 2-1: searching human face characteristic pixel points by artificial mucobacteria; the artificial mucobacteria walk around the face image, and a plurality of mucomorphs can search face characteristic pixel points in different directions and serve as external food sources to provide nutrition elements; further, the artificial mucobacteria adds the searched human face characteristic pixel points into the human face characteristic pixel point set of the mucoshape changing body, and updates the human face characteristic pixel point set of all the mucoshape changing bodies;
substep 2-2: extracting facial features by using artificial mucosae; the artificial mucobacteria adopt a knowledge-based characterization method when extracting the facial features, a facial knowledge base and a facial feature classification feature database are constructed according to the shape description of facial organs and the distance characteristics of different organs, and facial feature components are classified into local feature points such as facial contours, eyes, nose, mouth, chin and the like, euclidean distances, feature curvatures, angles and the like among the local feature points; the artificial mucobacteria describe the facial features by using the local organs of the facial and the geometric structure relations between the organs, so that the important geometric features of the facial can be accurately identified;
Substep 2-3: the mucoform body is continuously expanded; the sticky deformation body can be expanded continuously from the center of the face image to the periphery, and pixel points which accord with the face characteristics are continuously searched; the method comprises the steps that a deformable body learns human face characteristic pixel points on a human face image, positive feedback is formed on the human face characteristic pixel points by nutritional element values under the influence of learning factors, and the deformable body is gradually gathered on the human face characteristic pixel points; when the deformation body expands to the whole face image, the expansion process is finished, and the face feature vector of the deformation body is updated.
Step 3, the artificial mucobacteria match and identify the face image, which comprises three sub-steps:
substep 3-1: matching and calculating a face image; matching all the searched face feature pixel points with a face feature database by using the deformable body, and calculating a matching value by using a matching function; the face feature pixel points with better matching degree can obtain higher matching values, and the face feature pixel points with poorer matching degree obtain lower matching values;
substep 3-2: the artificial myxobacteria continuously shrink; the artificial mucobacteria starts to transport the nutrition elements on the searched face feature pixel points back into the human body, the matching values on the face feature pixel points start to be reduced, and the mucomorphs continuously consume the nutrition elements; the facial feature pixel points with lower matching values gradually disappear because the mucomorph absorbs nutrient elements or the matching values, namely the mucomorph contracts; the face feature pixel points with higher matching values can continuously provide nutrition elements or matching values, and the contracted deformable bodies can be gradually gathered under the action of learning factors, so that the connection of the deformable bodies is enhanced, and the matching values are improved;
Substep 3-3: judging the end of face recognition; further, a fixed iteration number can be preset, the preset iteration number is completed, and the face recognition is finished; further, a face feature pixel matching value may be preset, and when the matching value of the remaining face feature pixels is higher than the matching value, face recognition judgment may be ended.
And step 4, outputting an artificial mucoadhesive face recognition result, wherein the step comprises the following two substeps:
substep 4-1: outputting face images after face recognition; identifying the face by using obvious identification on the face identification image, and outputting a face image identification result with the identification;
substep 4-2: outputting relevant calculation parameters of the artificial myxobacteria; the matching values of all face characteristic pixel points comprise nutrition element values of the deformation body.
The substep 2-1: the artificial mucobacteria searching method for the human face characteristic pixel points comprises the following steps:
the artificial mucobacteria imitate biological foraging behaviors to expand a two-dimensional plane in a face image into a three-dimensional space, a plurality of mucodeformation bodies expand when the mucobacteria forages, the pictures are converted into the three-dimensional space for the mucobacteria forages, and the mucobacteria forages in the three-dimensional space to acquire three-dimensional feature vectors of the face; generating N-dimensional vector mapping in a three-dimensional face image, and putting M N-dimensional vectors into a face feature pixel point set S, wherein S= { Γ 1 ,Γ 2 ,Γ 3 ,……,Γ M }，Γ 1 ,Γ 2 ,Γ 3 ,……,Γ M Respectively isN-dimensional face feature point vector, and distance of N-dimensional face feature vectorx i 、y i Respectively an abscissa and an ordinate of the ith face feature pixel on the face image; artificial mucosae foraging is self-expanding and forms a feedback mechanism according to the distance d of the N-dimensional face feature vector 1 ,d 2 ,...,d N The learning factor of n pieces of deformation bodies in a three-dimensional space can be obtained as +.>Is a probability distribution vector of the mucosae search, +.>As the weight Γ i Is the eigenvector Γ i The size of the nutrient elements of (2); θ=θ 1 ,θ 2 ,...,θ i For all learner parameters, assume M image face feature vectors extracted based on the mucosae modelλ i Vector weight matrix for ith face feature vector,/->Learning factor phi for kth deformation in Mth face picture i The deviation vector is the i-th face feature pixel point; calculating the nutritional element values of the average face image after obtaining the probability distribution vector of the mucosae search +.>According to the learning factor I of artificial myxobacteria i The deviation vector of the ith face feature pixel point is calculated as phi i ＝I i - ψ, and thus the deviation vector of all n mucositive forms can be determined as Φ n The method comprises the steps of carrying out a first treatment on the surface of the Further, the artificial mucosae training model parameter θ= { λ, Φ }, kth (k=1, 2,3,..m) vector weight moment The matrix is->Locking a face coordinate frame L according to feature vectors of images i (i=1, 2,., M) and feature vector S 0 。
The substep 3-1: the face image matching calculation method comprises the following steps:
the face image matching calculation is used for obtaining the similarity of the face characteristic points; considering the unbalance of the samples, the artificial mucosae give higher weight to the samples matched with the model in the face feature database, thus deriving the expression of the weighting function as:wherein: m is the number of samples, N is the number of feature points matched with the model in the face feature database, and gamma n For different weights, |x|| is a distance vector between the matching feature points; since the sample weight is determined by the face image category such as: side face, mouth opening, head-up expression, etc., then γ n The corresponding function is: />The final weight expression is obtained, wherein for the kth variant, the ith face feature pixel point is +.>As the original weight, θ=θ 1 ,θ 2 ,...,θ i C is the number of different face categories for all learner parameters;
further, the face feature matching value calculating function is as follows:the degree of matching with the model in the face feature database can be calculated with reference to an algorithm commonly found in similarity or matching calculation; preferably, the matching value of the face feature is directly measured using euclidean distance:
The matching value calculation formula uses a distance definition and calculation method, can accurately measure the interrelationship of face feature organs, particularly the real distance between two points in a three-dimensional face or m-dimensional face space, and can also measure the natural length of a face feature vector, wherein the Euclidean distance in a multi-dimensional space is the real distance between two face feature pixel points, namely the distance in the three-dimensional space; and secondly, calculating cosine similarity, namely taking cosine values of two vector included angles in space as a measure for measuring the difference between individuals of two face feature pixel points: is a model of the abscissa of the pixel points of the face characteristics,a model of the ordinate of the pixel points of the face characteristics;
the line vectors formed by the model in the face feature database and the model of the vector formed by the key feature points extracted from the original face image are calculated by the two distance calculation methods respectively as follows: d (D) 1 ＝[d 11 ,d 12 ,d 13 ...d 1n ]、D 2 [d 21 ,d 22 ,d 23 ...d 2n ]Finally, calculating the similarity percentageThe matching value of the face image can be obtained.
The substep 3-2: the artificial myxobacteria continuously shrink, which comprises the following steps:
locking the food range in artificial mucosae foraging to shrink protoplasm, and keeping away from the area without food or the area with unobvious facial features; the mucoviscidosis uses the shrinkage characteristic of mucobacteria to make the facial feature vector S 0 Gradually adjusting the shape and finally gatheringTo the face pixel point with higher face feature matching value S; the deformable body continuously iterates to shrink the surrounding ring, each time shrink represents leaving the area without obvious face features, and normalizes the mucoid foraging face feature vector; further, the face feature points are expressed as vectors: s= [ x ] 1 ,y 1 ,x 2 ,y 2 ,...x k ,y k ] T Wherein x is i 、y i Respectively an abscissa and an ordinate of the ith face feature pixel on the face image; k is the number of feature points, the face outline and the key points of the five sense organs are regarded as mucosae, the mucosae foraging behavior is simulated to primarily find the food, then continuous learning and matching are carried out, the current food position is enabled to be continuously close to the real face feature point position in the learning and matching process, and the current food position is far away from the pixel points of the non-face features; preferably, a face regression learning function, i.e. a minimization function,wherein,a real face feature point vector representing the ith sample,/->The human face feature vector representing the current estimation of the ith sample, u is the regression adjustment quantity of each picture sample in the current cycle, L i Mould restriction, lambda, indicative of mucoid protoplast contraction i A vector weight matrix for the ith face feature vector; based on absolute deviation between the mucosae algorithm and the true value, the mucosae algorithm is calculated to obtain +. >Wherein: m is the number of face samples in the picture i to obtain a final target function of face image feature point matching and recognition +.>Wherein (1)>A real face feature point vector representing the ith sample,/->The human face feature vector representing the current estimation of the ith sample, u is the regression adjustment quantity of each picture sample in the current cycle, L i Mould restriction, lambda, indicative of mucoid protoplast contraction i A vector weight matrix for the ith face feature vector; the final shrinkage of the deformable body to the characteristic point coordinates (x) can be obtained according to the characteristic point objective function i ,y i )。
The invention discloses a human face recognition method based on artificial mucosae, which has the following technical effects:
1: aiming at the short plates with poor search algorithm performance in the current face recognition technology, the face recognition algorithm is improved. In combination with the existing face recognition algorithm, a face recognition algorithm based on biotechnology is provided. The algorithm can solve the problems that the computer identification is not deep enough and the interference of human factors is reduced, so that the face identification speed is faster, the accuracy is better, the identification is more accurate and the performance is better.
2: the method can efficiently complete the optimization solution of face recognition, simplify the training method through the learning training of biological characteristics, save social cost, improve the resource utilization efficiency and simplify the problem solving flow of researchers.
Drawings
FIG. 1 is a flowchart of an algorithm of the present invention.
Fig. 2 is an illustration of face recognition used in the test of the present invention.
Fig. 3 is a graph showing the result of the face recognition test of the present invention.
Detailed Description
FIG. 1 is a flowchart of an algorithm of the invention, a human mucoid-based face recognition method, including food sources, mucositis, in-vivo nutrient elements, and mucositis mass; the myxobacteria forages in a three-dimensional space, the myxobacteria protoplasm expands to form a mucoid body, an N-dimensional vector is formed, key points of facial features are regarded as food of the myxobacteria, and the myxobacteria continuously approximate to the positions of the real facial feature points through environmental training and learning; there are two modes of operation of the internal nutrient elements of the coliform bacteria, one is the weakening operation, namely the internal nutrient elements on all the coliform bodies are reduced in a certain ratio, and the process of absorbing and consuming the internal nutrient elements by the primordia in the real coliform bacteria in nature is simulated; the other is enhancement operation, namely the concentration of in-vivo nutrient elements is enhanced near the pixel nodes of the face image due to digestion and absorption, in positive feedback, the concentration distribution of the mucosae in the transportation of the nutrient elements in the space foraging network is continuously optimized, the ingestion network of the mucosae is continuously optimized, and the pixel point with the highest nutrient element is determined to be the final face characteristic pixel point through the orthogonal transformation of the mucosae ingestion sample formed by the food nutrient element ratio.
Step 1, initializing a face image and artificial mucobacteria, converting the image into a three-dimensional space for mucobacteria foraging, generating N-dimensional vector mapping by a mucoform in a flow network, and putting M N-dimensional vectors into a set S, wherein S= { Γ 1 ,Γ 2 ,Γ 3 ...Γ M N-dimensional vector distanceComprises three substeps:
substep 1-1: collecting a face image;
substep 1-2: preprocessing a face image;
substep 1-3: initializing artificial mucosae parameters; the colistin primordium feeds the human face in a three-dimensional space, the colistin deformation body generates colistin nonlinear search, the colistin searches for food and then carries out learning training, a human face characteristic pixel point network is formed, and the food is locked and nutrient elements are transported through distributed search and probabilistic nutrient element transportation. The training feedback of the mucosae and the positive feedback mechanism of the ingestion network enable the searching capability of the mucosae to the human face characteristics in the space to be further enhanced, and the accuracy of the human face identification of the mucosae is improved by increasing the comparison mode of the human face characteristics database.
And 2, extracting features of the human face image by using the artificial mucosae, wherein compared with the traditional data extraction, the invention mainly performs distributed search on the image through the expansion of the mucosae in foraging, and the original body of the mucosae is continuously expanded in the form of the mucosae in the foraging process, so that the mucosae are spread over all nodes in an image space. The positions of the mucobacteria are covered with the face pixels, and the face feature vectors can be extracted from the complex images according to a positive feedback learning mechanism formed by the face pixels. Comprises three substeps:
Substep 2-1: searching human face characteristic pixel points by artificial mucobacteria; since the foraging of the mucosae is a positive feedback mechanism formed by self-expansion, the mucosae n-stripe deformation learner in a three-dimensional space is a mucosae probability distribution vector. Assume that image face feature vectors are extracted based on slime moldAfter obtaining the mucositis probability distribution vector, calculating an average image, a deviation vector and finally locking a face pixel L according to the characteristic vector of the image i (i=1, 2,., M) and a eigenvalue vector S 0 。
Substep 2-2: extracting facial features by using artificial mucosae;
substep 2-3: the mucoform body is continuously expanded;
step 3, matching and identifying the face image by using the artificial mucor, wherein the step comprises three substeps:
substep 3-1: matching and calculating a face image;
substep 3-2: the artificial myxobacteria continuously shrink; when the myxobacteria ingest nutrient elements, the myxoshape is gradually adjusted, and the iteration of the myxoshape is used for narrowing the range, so that self-training is performed to find key points where food is located. Assuming that the facial contours and the facial features are used as food positions in the process, the myxobacteria are continuously trained through continuous training and the deformable body pipeline is continuously and iteratively contracted, a network only comprising facial feature pixel points is finally formed, the obtained facial feature pixel points are used as facial recognition key points for coordinated processing, thus obtaining facial key point feature value vectors composed of the key points, Obtaining the final objective function of face image feature point matching and recognitionWherein (1)>A real face feature point vector representing the ith sample,/->The human face feature vector representing the current estimation of the ith sample, u is the regression adjustment quantity of each picture sample in the current cycle, L i Mould restriction, lambda, indicative of mucoid protoplast contraction i A vector weight matrix for the ith face feature vector; the final shrinkage of the deformable body to the characteristic point coordinates (x) can be obtained according to the characteristic point objective function i ,y i )。
Substep 3-3: judging the end of face recognition; after face features are extracted and face key points are matched based on a mucosae foraging behavior process, matching values among the face key points are calculated, and a matching quantification result is obtained to judge whether the face recognition process is finished. The processing procedures of the substep 3-2 and the substep 3-3 project the face feature data obtained in the step 2 through the iterative training process of the deformation body in a subspace to obtain a calculated value capable of measuring the matching degree and the accuracy of the face recognition result. The step can judge whether to finish the face recognition according to the matching value, or judge whether to finish the face recognition according to the iteration step.
Step 4, outputting the human face recognition result of the artificial mucosae; for the similarity calculation of the geometry of the human face, euclidean measurement or cosine similarity measurement is usually adopted, and the human face characteristic points obtained in the artificial mucosae ingestion process can be influenced by environmental factors to cause errors, so that a weighting function expression can be introduced for correction when the matching degree or similarity of the human face characteristic is calculated; preferably, cosine similarity is adopted to accurately evaluate the similarity of the facial feature vectors. The match end comparison sub-steps are as follows: calculating the original faceCosine similarity between each vector extracted by training the image and the face to be compared, and respectively storing the obtained data in two column vectors D 1 、D 2 In (a) and (b); defining a variable eta for measuring the similarity, wherein the value of the variable eta is a vector D 1 、D 2 The average value of the element ratio sum isThis variable is referred to as a face match value or degree of match. Comprises two substeps: substep 4-1: outputting face images after face recognition; substep 4-2: outputting relevant calculation parameters of the artificial myxobacteria;
examples:
in the embodiment of the invention, a bionic optimization method for face recognition problem based on myxobacteria foraging behavior myxose body expansion and contraction training is provided, and as shown in fig. 2, the images with faces used for testing are shown in the invention, and are derived from test databases of image research at home and abroad; as shown in fig. 3, the image of the result of the face recognition test of the present invention is that the red frame has been used to identify the face;
Step 1, initializing a face image and artificial mucobacteria, sampling the image through mucobacteria foraging behaviors, marking n points in the image as training data after obtaining shape vectors and characteristic point key information, wherein the marking positions of the key points can represent target food, the selected key marks are arranged on the outline and the edge of the mucobacteria food, and the vectors of the marking points are expressed as follows: s is S i ＝(x i1 ,y i1 ,x i2 ,y i2 ...x in ,y in ) T Wherein (x) ij ,y ij ) Represents the coordinates of the j-th contour point in the i-th image, and n represents the number of mark points in each image. S represents a shape vector, and feature information near a food key point is acquired. Comprises three substeps:
substep 1-1: collecting a face image; as shown in fig. 2, the facial image is derived from a database.
Substep 1-1a: setting space food contour points, sampling the image, obtaining shape vectors and key information of feature points, and marking n points as training data.
Substep 1-1b: setting S to represent a shape vector, (x) ij ,y ij ) Representing the coordinates of the j-th contour point in the i-th drawing.
Substep 1-1c: calculating N-dimensional vector distance
Substep 1-2: preprocessing a face image;
substep 1-2a: will feature vector S 0 Normalization.
Substep 1-2b: the vector form representing the face feature points is S= [ x ] 1 ,y 1 ,x 2 ,y 2 ,...x k ,y k ] T Wherein x, y are coordinates of the feature points; k is the number of feature points.
Substep 1-3: initializing artificial mucosae parameters;
substep 1-3a: the vector of set keypoints is expressed as: s is S i ＝(x i1 ,y i1 ,x i2 ,y i2 ...x in ,y in ) T 。
Substep 1-3b: computing regression learning minimization function based on characteristics of slime mold
And 2, extracting characteristics of the human face image by using the artificial mucobacteria, expanding protoplasmic clusters when the mucobacteria forge, enabling the mucobacteria to perform distributed search in different directions, regarding the image as a three-dimensional space, enabling the mucobacteria to forge in the space to cover all nodes of the space as much as possible, enabling the selection and the route of the mucobacteria to be probabilistic, and forming a specific n-dimensional vector distribution network in a human face frame to obtain human face characteristic vectors. Comprises three substeps:
substep 2-1: searching human face characteristic pixel points by artificial mucobacteria;
substep 2-1a: calculating the results of the myxobacteria n deformation learner in the three-dimensional space as myxobacteria probability distribution vectors As the weight, θ=θ 1 ,θ 2 ,...,θ i For all learner parameters.
Substep 2-1b: calculating the average image after calculating the slime probability distribution vectorCalculating the deviation vector as phi i ＝I i - ψ, trained model parameters θ= { λ, Φ }.
Substep 2-2: extracting facial features by using artificial mucosae;
substep 2-2a: m image face feature vectors extracted by using mucobacteria model are calculated
Substep 2-2b: the calculation of the k (k=1, 2,3, M) vector weight matrices as
Substep 2-2c: projecting a sample for training the face image into a subspace through a projection matrix W, and calculating to obtain a characteristic value of the subspace and a face characteristic value: y is i ＝W T x i ,i＝1,2...C。
Substep 2-2d: calculating according to the substeps 3-2a,3-2b and 3-2c to obtain the corresponding food principal component feature vector W= [ W ] 1 ,w 2 ...w d ],d<<M。
Substep 2-2e: the formula for calculating the average value of the nutrition elements of the colistin food is as follows:
substep 2-2f: the covariance matrix of the eigenvalue is calculated as follows:
substep 2-3: the mucoform body is continuously expanded;
substep 2-3a: put M N-dimensional vectors into set S, and initially set the vector to map to set S= { Γ 1 ,Γ 2 ,Γ 3 ,……,Γ M }。
Substep 2-3b: obtaining the feature vector locking face coordinate frame L of the image i (i=1, 2,., M) and feature vector S 0 。
Step 3, matching and identifying the human face image by using the artificial mucosae, further searching food, adjusting vector mapping formed by the mucosae step by step, iteratively narrowing the range of the mucosae, and performing continuous iteration, wherein the mucosae are self-trained and find food key points, the human face contour and the five sense organ key points are regarded as food of the artificial mucosae, and the current food position is enabled by the mucosae to be continuously approximate to the real human face characteristic point position through iterative training and learning; comprises three substeps:
Substep 3-1: matching and calculating a face image;
substep 3-1a: calculating the shape of each food sample, and averaging to obtain an average shape vector
Substep 3-1b: face shape vector matrix
Substep 3-1c: arranging the characteristic values in descending order, and selecting t large characteristic values
Substep 3-1d: the calculation has absolute deviation from the true value, and the calculationWhere m is the number of face samples within picture i.
Substep 3-1e: calculating final characteristic point objective function
Substep 3-2: the artificial myxobacteria continuously shrink;
substep 3-2a: calculating a weighting functionWherein M is the number of samples, N is the number of feature points, and gamma n For the different weights to be used, the distance vector between feature points.
Substep 3-2b: according to the face image category factors, calculating the final weight gamma n Corresponding function
Substep 3-2c: a weight refinement function is calculated and the weight refinement function,
substep 3-2d: calculating an n-dimensional spatial two-point Euclidean distance metric
Substep 3-2e: calculating a cosine value of a vector angle of a magnitude measure spatially measuring a difference between two individuals
Substep 3-3: judging the end of face recognition;
substep 3-3a: calculating the feature vector of the original face image and the face database, and extracting the module row vector D of the vector formed by the key feature points 1 ＝[d 11 ,d 12 ,d 13 ...d 1n ]、D 2 [d 21 ,d 22 ,d 23 ...d 2n ]。
Substep 3-3b: calculating the similarity percentage of face recognitionAnd obtaining the accuracy of the face recognition result.
Step 4, outputting the human face recognition result of the artificial mucosae; the mucosae ingest food nutrition through repeated iteration, and finally, the mucosae gradually shrink on the key point food with higher nutrient elements in the space; the mucosae body vectors have different proportional sizes due to the influence of the nutrition elements, the feeding network of the mucosae body is optimized by the change of the feeding network nutrition element transportation concentration in the mucosae learning space, orthogonal transformation on mucosae feeding samples is formed through the enrichment degree of the food nutrition elements, final face feature pixel points are determined, and the face feature vectors and the face feature matching values of the mucosae body are obtained. Comprises two substeps:
substep 4-1: outputting face images after face recognition; as indicated by the red boxes in fig. 3. Food is further searched by locking the range of food in mucosae foraging and carrying out protoplasm contraction, and the algorithm utilizes the biological characteristics of mucosae to carry out facial feature vector S 0 And performing step-by-step shape adjustment to finally obtain the estimation S of the face characteristic points. As shown by the red boxes in fig. 3, the faces in the image are accurately recognized.
Substep 4-2: outputting relevant calculation parameters of the artificial myxobacteria; the colistin primordial group forges food in a three-dimensional space, the mucomorph body outputs a human face feature vector, a learning training result of the colistin for searching food and a human face food source pixel point network.
Claims (8)
1. The human face recognition method based on the artificial myxobacteria is characterized by comprising the following steps of: matching and identifying the facial image by simulating the feeding behavior of the mucosae, and extracting facial feature vectors through the expansion and contraction processes of the artificial mucosae in the facial image; simulating the initial position of a single mucor as a center pixel point which is the center of a face image, and simulating the face characteristic pixel point as an external food source; comparing and matching the extracted face feature vector with a face feature model stored in a face feature database, wherein the size of the matching value can be simulated into the size of an external food source, namely the larger the face matching value is, the larger the food source is, otherwise, the smaller the face matching value is, the smaller the food source is; simulating a virtual edge between two pixel points in the face feature vector as a deformation body, and simulating a pixel point difference value on the deformation body as a nutrition element value; simulating the expansion behavior of the deformable body as searching for the facial feature vector, and simulating the contraction of the deformable body as matching for the facial feature vector;
according to the human face recognition method based on the artificial mucosae, the facial images are matched and recognized by simulating mucosae foraging behaviors, and the facial feature vectors are extracted through the expansion and contraction processes of the artificial mucosae in the facial images; the method comprises the following steps:
Step 1, initializing a face image and artificial mucobacteria;
collecting a face image, preprocessing the face image, performing light compensation, gray level transformation, histogram equalization, normalization, geometric correction, filtering and sharpening on each pixel point of the face image, and constructing a face image preprocessing vector according to a preprocessing result; initializing the artificial mucobacteria, fixing the cell nucleus of the artificial mucobacteria at the center point of the face image, and initializing the quantity of the mucoshape bodies of the artificial mucobacteria by using a random number; setting a human face characteristic pixel point set in the deformable body as an empty set, wherein human face characteristic vectors formed by all human face characteristic pixel point sets are empty sets, namely the deformable body does not find human face characteristic pixel points or external food sources during initialization; removing the values of nutrient elements in the visco-deformation body, wherein the transport of the nutrient elements is not carried out in the visco-deformation body temporarily; after the initialization is completed, entering a step 2;
step 2, extracting characteristics of the human face image by using artificial mucobacteria:
expanding the mucomorph body of the artificial mucobacteria on the face image everywhere, searching face feature pixel points, adding the searched face feature pixel points into a face feature pixel point set of the mucomorph body, updating the face feature pixel point set of all the mucomorph bodies, and updating face feature vectors; the artificial mucobacteria are used for extracting the human face characteristics, the mucomorph body can be connected with different human face characteristic pixel points for nutrient element transportation, the external food source is large or the matching value of the human face characteristic pixel points is high, more nutrient elements can be provided, the external food source is small or the nutritional elements provided by the low matching value of the human face characteristic pixel points are also less, and the nutrient element values in the mucomorph body are updated; the mucomorph body is continuously expanded until all face feature pixel points are found, the artificial mucobacteria expansion process is finished, the face feature extraction is finished, and the step 3 is entered;
Step 3, matching and identifying the face image by using artificial mucobacteria:
expanding the mucomorph of the artificial mucobacteria around the face image, searching the face characteristic pixel points, and adding the searched face characteristic pixel points into a face characteristic pixel point set of the mucomorph; then, the artificial mucobacteria start to continuously shrink, namely, the external food sources or the human face characteristic pixel points are digested, so that the human face matching and recognition functions are realized;
in the step, the artificial mucobacteria start to continuously shrink, namely, the external food sources or the human face characteristic pixel points are digested, so that the human face matching and recognition functions are realized; the mucomorph starts to absorb the nutrition elements on the external food source or the human face characteristic pixel points, and the simulation real mucobacteria transport the nutrition elements into the human body, so that the matching values of the nutrition elements on the external food source or the human face characteristic pixel points start to be reduced; the face characteristic pixel points with low matching values can be absorbed with nutrition elements or matching values to disappear, and the face characteristic pixel points with high matching values can provide nutrition elements or matching values to strengthen; updating the nutrition element values in the sticky deformation body and updating the matching values on the face characteristic pixel points; the mucosae body continuously contracts until the matching values on all the face characteristic pixel points accord with a preset threshold value, the artificial mucobacteria contraction process is finished, the face matching and recognition process is finished, and the step 4 is entered;
Step 4, outputting the human face recognition result of the artificial colistin:
outputting face images after face recognition, identifying faces in the images, outputting a matched and recognized face characteristic pixel point matrix, and outputting related calculation parameters of artificial mucobacteria, wherein the related calculation parameters comprise nutrition element values of mucomorphs, and matching values of all face characteristic pixel points.
2. The human-based myxobacteria face recognition method of claim 1, wherein the method comprises the steps of:
the step 1, the initialization of the face image and the artificial mucosae, comprises three substeps:
substep 1-1: collecting a face image; different face images are acquired by using various shooting devices, including a digital camera for acquiring a static face image, a video camera for acquiring a dynamic face image, a camera array for acquiring face images of different positions and angles, and a video camera with different focuses for simultaneously acquiring different expressions of the face; when a user enters the acquisition range of the shooting equipment, the shooting equipment automatically searches and acquires face images of the user; after finishing face image acquisition in the substep 1-1, turning to the substep 1-2 for pretreatment;
substep 1-2: preprocessing a face image; the method mainly comprises the steps of light compensation, gray level transformation, histogram equalization, image normalization, geometric correction, image filtering and sharpening of a face image; different preprocessing functions are adopted to eliminate deviation and noise existing in the face image acquired in the substep 1-1, so that the quality of the face image is improved; initializing a face feature database, and initializing a face feature matching function for face feature matching and matching value calculation;
Substep 1-3: initializing artificial mucosae parameters; setting cell nuclei of the artificial mucobacteria on a central pixel point of a face image, setting the quantity of mucoshape bodies of the artificial mucobacteria, setting learning factors of the artificial mucobacteria, eliminating nutrition element values in the mucoshape bodies to be 0, eliminating face characteristic pixel point sets in the mucoshape bodies to be empty sets, and eliminating face characteristic vectors to be 0; and importing a face feature database and a face feature model.
3. The human-based myxobacteria face recognition method of claim 1, wherein the method comprises the steps of:
step 2, the characteristic extraction of the human face image by the artificial mucosae comprises three substeps:
substep 2-1: searching human face characteristic pixel points by artificial mucobacteria; the artificial mucobacteria walk around the face image, and a plurality of mucomorphs can search face characteristic pixel points in different directions and serve as external food sources to provide nutrition elements; further, the artificial mucobacteria adds the searched human face characteristic pixel points into the human face characteristic pixel point set of the mucoshape changing body, and updates the human face characteristic pixel point set of all the mucoshape changing bodies;
substep 2-2: extracting facial features by using artificial mucosae; the artificial mucobacteria adopt a knowledge-based characterization method when extracting the facial features, a facial knowledge base and a facial feature classification feature database are constructed according to the shape description of facial organs and the distance characteristics of different organs, and facial feature components are classified into facial contours, eyes, nose, mouth and chin local feature points, euclidean distances, feature curvatures and angles among the local feature points; the artificial mucobacteria describe the facial features by using the local organs of the human face and the geometric structure relation between the organs, so that the important geometric features of the human face can be accurately identified;
Substep 2-3: the mucoform body is continuously expanded; the sticky deformation body can be expanded continuously from the center of the face image to the periphery, and pixel points which accord with the face characteristics are continuously searched; the method comprises the steps that a deformable body learns human face characteristic pixel points on a human face image, positive feedback is formed on the human face characteristic pixel points by nutritional element values under the influence of learning factors, and the deformable body is gradually gathered on the human face characteristic pixel points; when the deformation body expands to the whole face image, the expansion process is finished, and the face feature vector of the deformation body is updated.
4. The human-based myxobacteria face recognition method of claim 1, wherein the method comprises the steps of:
step 3, the artificial mucobacteria match and identify the face image, which comprises three sub-steps:
substep 3-1: matching and calculating a face image; matching all the searched face feature pixel points with a face feature database by using the deformable body, and calculating a matching value by using a matching function; the face feature pixel points with good matching degree can obtain a high matching value, and the face feature pixel points with poor matching degree have a low matching value;
substep 3-2: the artificial myxobacteria continuously shrink; the artificial mucobacteria starts to transport the nutrition elements on the searched face feature pixel points back into the human body, the matching values on the face feature pixel points start to be reduced, and the mucomorphs continuously consume the nutrition elements; the facial feature pixels with low matching values gradually disappear because the mucomorph absorbs nutrient elements or the matching values, namely the mucomorph contracts; the face feature pixel points with high matching values can continuously provide nutrition elements or matching values, and the contracted mucomorphs are gradually gathered under the action of learning factors, so that the connection of the mucomorphs is enhanced, and the matching values are improved;
Substep 3-3: judging the end of face recognition; presetting a fixed iteration number, finishing the preset iteration number, and finishing the face recognition judgment; or presetting a face characteristic pixel point matching value, and ending face recognition judgment when the matching value of the rest face characteristic pixel points is higher than the value.
5. The human-based myxobacteria face recognition method of claim 1, wherein the method comprises the steps of:
and step 4, outputting an artificial mucoadhesive face recognition result, wherein the step comprises the following two substeps:
substep 4-1: outputting face images after face recognition; identifying the face by using obvious identification on the face identification image, and outputting a face image identification result with the identification;
substep 4-2: outputting relevant calculation parameters of the artificial myxobacteria; the matching values of all face characteristic pixel points comprise nutrition element values of the deformation body.
6. A human mucosae-based face recognition method according to claim 3, wherein:
the substep 2-1: the artificial mucobacteria searching method for the human face characteristic pixel points comprises the following steps:
artificial mucobacteria simulate biological foraging behavior to image human faceExpanding the two-dimensional plane in the plane (B) to a three-dimensional space, expanding a plurality of mucodeformation bodies when mucobacteria forge, converting the picture into the three-dimensional space for mucobacteria forge, and allowing mucobacteria to forge in the three-dimensional space to acquire a three-dimensional feature vector of a human face; generating N-dimensional vector mapping in a three-dimensional face image, and putting M N-dimensional vectors into a face feature pixel point set S, wherein S= { Γ 1 ,Γ 2 ,Γ 3 ,……,Γ M }，Γ 1 ,Γ 2 ,Γ 3 ,……,Γ M Respectively N-dimensional face feature point vectors and distances of the N-dimensional face feature vectorsx i 、y i Respectively an abscissa and an ordinate of the ith face feature pixel on the face image; artificial mucosae foraging is self-expanding and forms a feedback mechanism according to the distance d of the N-dimensional face feature vector 1 ,d 2 ,...,d N The learning factors of n pieces of deformation bodies in a three-dimensional space are obtained as follows: />Is a probability distribution vector of the mucosae search, +.>Is the weight +. i I is the feature vector Γ i The size of the nutrient elements of (2); θ=θ 1 ,θ 2 ,...,θ i For all of the learner parameters,
m image face feature vectors extracted based on myxobacteria model are setλ i Vector weight matrix for ith face feature vector,/->Learning factor phi for kth deformation in Mth face picture i Deviation direction of ith face feature pixel pointAn amount of; calculating the nutritional element values of the average face image after obtaining the probability distribution vector of the mucosae search +.>According to the learning factor I of artificial myxobacteria i The deviation vector of the ith face feature pixel point is calculated as phi i ＝I i - ψ, and then find the deviation vector of all n mucositions as Φ n The method comprises the steps of carrying out a first treatment on the surface of the Artificial mucosae training model parameter θ= { λ, Φ }, kth vector weight matrix is +.>k=1, 2,3, M; locking a face coordinate frame L according to feature vectors of images i And feature vector S 0 Wherein i=1, 2,..m.
7. The human-based myxobacteria face recognition method of claim 4, wherein the human-based myxobacteria face recognition method is characterized by:
the substep 3-1: the face image matching calculation method comprises the following steps:
the face image matching calculation is used for obtaining the similarity of the face characteristic points; considering the unbalance of the samples, the artificial mucosae give higher weight to the samples matched with the model in the face feature database, thus deriving the expression of the weighting function as:wherein: m is the number of samples, N is the number of feature points matched with the model in the face feature database, and gamma n For different weights, |x|| is a distance vector between the matching feature points; since the sample weights are subject to the face image categories: side face, mouth opening, head-up expression factor influence, gamma n The corresponding function is: />The final weight expression is obtained, wherein for the kth variant, the ith face feature pixel point，/>As the original weight, θ=θ 1 ,θ 2 ,...,θ i C is the number of different face categories for all learner parameters;
the face feature matching value calculating function is as follows:the degree of matching with the model in the face feature database can be calculated with reference to an algorithm commonly found in similarity or matching calculation; matching values of face features are directly measured using euclidean distances: / >
The matching value calculation formula uses a distance definition and calculation method, can accurately measure the interrelation of face feature organs, calculates the real distance between two points in a three-dimensional face or m-dimensional face space, and can measure the natural length of a face feature vector, wherein the Euclidean distance in a multidimensional space is the real distance between two face feature pixel points, namely the distance in the three-dimensional space; and secondly, calculating cosine similarity, namely taking cosine values of two vector included angles in space as a measure for measuring the difference between individuals of two face feature pixel points: the module of the abscissa of the pixel points with the characteristics of the human face is +.>A model of the ordinate of the pixel points of the face characteristics;
calculating the model in the human face characteristic database and the original human face image by the two distance calculation methodsThe row vectors formed by the modes of the vectors formed by the extracted key feature points are respectively as follows: d (D) 1 ＝[d 11 ,d 12 ,d 13 ...d 1n ]、D 2 [d 21 ,d 22 ,d 23 ...d 2n ]Finally, calculating the similarity percentageAnd obtaining the matching value of the face image.
8. The human-based myxobacteria face recognition method of claim 4, wherein the human-based myxobacteria face recognition method is characterized by:
the substep 3-2: the artificial myxobacteria continuously shrink, which comprises the following steps:
locking the food range in artificial mucosae foraging to shrink protoplasm, and keeping away from the area without food or the area with unobvious facial features; the mucoviscidosis uses the shrinkage characteristic of mucobacteria to make the facial feature vector S 0 Gradually performing shape adjustment, and finally gathering to face pixel points with high face feature matching values S; the deformable body continuously iterates to shrink the surrounding ring, each time shrink represents leaving the area without obvious face features, and normalizes the mucoid foraging face feature vector; the face feature points are expressed as vectors: s= [ x ] 1 ,y 1 ,x 2 ,y 2 ,...x k ,y k ] T Wherein x is i 、y i Respectively an abscissa and an ordinate of the ith face feature pixel on the face image; k is the number of feature points, the face outline and the key points of the five sense organs are regarded as mucosae, the mucosae foraging behavior is simulated to primarily find the food, then continuous learning and matching are carried out, the current food position is enabled to be continuously close to the real face feature point position in the learning and matching process, and the current food position is far away from the pixel points of the non-face features; a face regression learning function, i.e. a minimization function,
wherein,a real face feature point vector representing the ith sample,/->The human face feature vector representing the current estimation of the ith sample, u is the regression adjustment quantity of each picture sample in the current cycle, L i Mould restriction, lambda, indicative of mucoid protoplast contraction i A vector weight matrix for the ith face feature vector; based on absolute deviation between the mucosae algorithm and the true value, the mucosae algorithm is calculated to obtain +. >Wherein: m is the number of face samples in the picture i to obtain a final target function of face image feature point matching and recognition +.>
Wherein,a real face feature point vector representing the ith sample,/->The human face feature vector representing the current estimation of the ith sample, u is the regression adjustment quantity of each picture sample in the current cycle, L i Mould restriction, lambda, indicative of mucoid protoplast contraction i A vector weight matrix for the ith face feature vector; the final shrinkage of the deformable body to the characteristic point coordinates (x) can be obtained according to the characteristic point objective function i ,y i )。
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202110977206.XA CN113743266B (en) | 2021-08-24 | 2021-08-24 | Human face recognition method based on artificial myxobacteria |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202110977206.XA CN113743266B (en) | 2021-08-24 | 2021-08-24 | Human face recognition method based on artificial myxobacteria |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN113743266A CN113743266A (en) | 2021-12-03 |
| CN113743266B true CN113743266B (en) | 2023-11-24 |
Family
ID=78732639
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202110977206.XA Active CN113743266B (en) | 2021-08-24 | 2021-08-24 | Human face recognition method based on artificial myxobacteria |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN113743266B (en) |
Families Citing this family (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN115879776B (en) * | 2023-03-02 | 2023-06-06 | 四川宏华电气有限责任公司 | Dangerous area early warning method and system applied to petroleum drilling machine |
Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN105069445A (en) * | 2015-07-09 | 2015-11-18 | 湖南工业大学 | Face recognition method based on improved bacterial foraging algorithm |
| CN110288131A (en) * | 2019-06-04 | 2019-09-27 | 三峡大学 | Transportation network fragility problem bionic optimization method based on slime mould foraging behavior |
| CN110298491A (en) * | 2019-06-04 | 2019-10-01 | 三峡大学 | A kind of transportation network node selecting method based on slime mould swarm intelligence |
Family Cites Families (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US9552374B2 (en) * | 2013-08-19 | 2017-01-24 | Kodak Alaris, Inc. | Imaging workflow using facial and non-facial features |
-
2021
- 2021-08-24 CN CN202110977206.XA patent/CN113743266B/en active Active
Patent Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN105069445A (en) * | 2015-07-09 | 2015-11-18 | 湖南工业大学 | Face recognition method based on improved bacterial foraging algorithm |
| CN110288131A (en) * | 2019-06-04 | 2019-09-27 | 三峡大学 | Transportation network fragility problem bionic optimization method based on slime mould foraging behavior |
| CN110298491A (en) * | 2019-06-04 | 2019-10-01 | 三峡大学 | A kind of transportation network node selecting method based on slime mould swarm intelligence |
Non-Patent Citations (1)
| Title |
|---|
| a node selecting approach for traffic network based on artificial slime mold;ZHENGYING CAI 等;《IEEEAccess》;第8436-8448页 * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN113743266A (en) | 2021-12-03 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN108615010B (en) | Facial expression recognition method based on parallel convolution neural network feature map fusion | |
| CN108182427B (en) | Face recognition method based on deep learning model and transfer learning | |
| CN112560631B (en) | Knowledge distillation-based pedestrian re-identification method | |
| CN107066559B (en) | Three-dimensional model retrieval method based on deep learning | |
| CN111611924B (en) | Mushroom identification method based on deep migration learning model | |
| CN105608471A (en) | Robust transductive label estimation and data classification method and system | |
| CN104143076B (en) | The matching process of face shape and system | |
| CN106980848A (en) | Facial expression recognizing method based on warp wavelet and sparse study | |
| CN108830289B (en) | Image clustering method and device based on improved fuzzy C-means clustering | |
| CN110717451B (en) | Medicinal plant leaf disease image identification method based on deep learning | |
| US20210117648A1 (en) | 3-dimensional model identification | |
| CN107862680B (en) | Target tracking optimization method based on correlation filter | |
| Beksi et al. | Object classification using dictionary learning and rgb-d covariance descriptors | |
| CN106250918B (en) | A kind of mixed Gauss model matching process based on improved soil-shifting distance | |
| Wang et al. | Research on face recognition technology based on PCA and SVM | |
| CN106845462A (en) | The face identification method of feature and cluster is selected while induction based on triple | |
| Fan et al. | Smoothness-driven consensus based on compact representation for robust feature matching | |
| CN113743266B (en) | Human face recognition method based on artificial myxobacteria | |
| Xu et al. | Generalized robust PCA: A new distance metric method for underwater target recognition | |
| Zhang et al. | Contrastive learning ideas in underwater terrain image matching | |
| CN111563180A (en) | Trademark image retrieval method based on deep hash method | |
| Padala et al. | Effect of input noise dimension in GANs | |
| CN113723482B (en) | Hyperspectral target detection method based on multi-example twin network | |
| Si et al. | Age-invariant face recognition using a feature progressing model | |
| Chun-Rong | Research on face recognition technology based on deep learning |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant | ||
| TR01 | Transfer of patent right |
Effective date of registration: 20241014 Address after: 518000 1104, Building A, Zhiyun Industrial Park, No. 13, Huaxing Road, Henglang Community, Longhua District, Shenzhen, Guangdong Province Patentee after: Shenzhen Hengyuan Zhida Information Technology Co.,Ltd. Country or region after: China Address before: 443002 No. 8, University Road, Xiling District, Yichang, Hubei Patentee before: CHINA THREE GORGES University Country or region before: China |