Research On Shape Recognition Algorithm For Fish Image Of Dynamic Deformation

Shape matching is a hot issue in computer vision, it is a fundamental problem in automatic image recognition and understanding, and has widely apply to object recognition, classification, content based image retrieval and so on. Although lots of studies have done, but until now there is not good method about shape matching of non-rigid objects deformation. For images acquired from real-time detection system, dynamic deformation inevitably occurs on the shape of object. In this situation, it is difficult to select classification feature. As deformation usually occurred on fish images from real-time detection system, a new recognition algorithm is presented in this paper.First of all, color image of fish should be processed to skeleton image before correcting, and the directional information of skeleton is obtained by Hough transform method, in the meantime, the cutting point needed for correcting is also obtained. Then corrected binary image is obtained after image cutting, image rotating and image merging. Corrected image has similar mapping image with standard fish shape by even-grid-polar mapping algorithm, polar mapping algorithm ensures the invariant in translation, rotation and scaling. local extreme point value and distance of adjacent extreme points (relative position of extreme points) as two features for recognition are extracted, and the matching degree by using matching recognition algorithm are calculated. Experimental results show that corrected image has a good matching degree, and the algorithm improves recognition effect against fish image which had dynamic deformation feature.The novelty of the paper is as follows:1. In deformation correcting algorithm, lines are extracted with Hough transform method, this could be used to determine cutting point effectively. Then image cutting, image rotating and image merging are used for correcting deformation of fish image; 2. Relative position information of extreme points from polar mapping image is extracted as the second feature, this could increase recognition accuracy, and make accurate classification for fish image, and improve the effectiveness of the algorithm.