| CPC G06V 40/10 (2022.01) [G06V 10/24 (2022.01); G06V 10/267 (2022.01); G06V 10/44 (2022.01); G06V 10/70 (2022.01); G06V 10/751 (2022.01); G06V 10/761 (2022.01)] | 2 Claims |
|
1. A dog nose print recognition method, comprising: collecting a nose image of a dog, acquiring the nose image, and processing the nose image to obtain a plurality of regional images to be recognized;
performing key point detection on the plurality of regional images to be recognized to obtain key points corresponding to the regional images to be recognized, and using the key points to perform alignment processing of the regional images to be recognized to obtain aligned regional images to be recognized; and
performing dog nose print feature vector extraction and recognition on the aligned regional images to be recognized, and determining a dog identity recognition result through the dog nose print feature vector extraction and recognition,
wherein the step of processing the nose image to obtain a plurality of regional images to be recognized comprises:
performing an edge contour process to the nose image to obtain a nose overall region bounded by a nose contour;
performing an image region division process to the nose overall region to obtain the plurality of regional images to be recognized, comprising:
recognizing and marking contour lines of a left nostril and a right nostril in the nose overall region to obtain a position of a contour line of the left nostril and a position of a contour line of the right nostril, and a center line between the left nostril and the right nostril, and taking the center line between the left nostril and the right nostril as a center line of the nose overall region;
lining a highest point of the contour line of the left nostril with a highest point of the contour line of the right nostril to obtain a horizontal higher line, and extracting a position of an intersection point of the horizontal higher line and the center line of the nose overall region as a first intersection point position;
extracting a position of an intersection point of the center line of the nose overall region and an upper contour line of the nose overall region as a second intersection point position;
lining the highest point of the contour line of the left nostril and the highest point of the contour line of the right nostril with the second intersection point position respectively, to respectively obtain an included angle between a line connecting the highest point of the contour line of the left nostril with the second intersection point position and the horizontal higher line, and an included angle between a line connecting the highest point of the contour line of the right nostril with the second intersection point position and the horizontal higher line;
utilizing the included angles and a center point-included angle determination model to obtain a position of a center point on the center line of the nose overall region; simultaneously, utilizing a distance between the first intersection point position and the second intersection point position in conjunction with a radius determination model to obtain a radius, wherein the center point-included angle determination model and the radius determination model are as follows respectively:
 wherein V represents an included angle between a line connecting the center point with the highest point of the contour line of the left nostril or the highest point of the contour line of the right nostril and the horizontal higher line; α1 and α2 respectively represent the included angle between the line connecting the highest point of the contour line of the left nostril with the second intersection point position and the horizontal higher line, and the included angle between the line connecting the highest point of the contour line of the right nostril with the second intersection point position and the horizontal higher line; α0 represents a benchmark comparison value of included angle with a value range of 17.8°-25.2°; and if α1 and α2 are equal, then (min(α1, α2)−α0)/(max(α1, α2)−α0)=0.1; R represents the radius; L represents the first the distance between the first intersection point position and the second intersection point position;
taking the center point as a circle center to form a first region to be recognized based on the radius, and utilizing the first region to be recognized to intercept a first regional image to be recognized in the nose overall region;
lining a lowest point of the contour line of the left nostril with a lowest point of the contour line of the right nostril to obtain a horizontal lower line, and extracting a position of an intersection point of the horizontal lower line and the center line of the nose overall region as a third intersection point position;
extracting a shortest distance between the lowest point of the contour line of the left nostril and the center line of the nose overall region and a shortest distance between the lowest point of the contour line of the right nostril and the center line of the nose overall region, respectively; taking the third intersection point position as a circle center, taking the shortest distance between the lowest point of the contour line of the left nostril and the center line of the nose overall region as a major axis and taking the shortest distance between the lowest point of the contour line of the right nostril and the center line of the nose overall region as a minor axis, to form a longitudinal elliptical region, utilizing the longitudinal elliptical region as a second region to be recognized, and utilizing the second region to be recognized to intercept a second regional image to be recognized in the nose overall region;
wherein the step of performing dog nose print feature vector extraction and recognition on the aligned regional images to be recognized, and determining a dog identity recognition result through the dog nose print feature vector extraction and recognition further comprises:
inputting the aligned regional images to be recognized into a trained feature extraction model to extract a dog nose print feature vector, to obtain a nose print feature vectors of a dog to be tested;
comparing the nose print feature vector of the dog to be tested with a reference dog nose print feature vector stored in a preset dog nose print feature vector library in terms of distance, and taking the reference dog nose print feature vector with closest distance to the nose print feature vector of the dog to be tested as a target dog nose print feature vector;
selecting the target dog nose print feature vector, and retrieving a dog identity information corresponding to the target dog nose print feature vector from the dog nose print feature vector library.
|