The Research On Algorithms For Recognition Of Cow Image Based On Wavelet Moment

Cow recognition is a main part of cow management. The cow recognition was relied on manual recognition in the past. The electrical labels are used on the cow recognition recently. It is a new challenge that the cow is recognized by analyzing the cow images.The recognition methods presented by the thesis were separated into 3 steps which are removing the illumination affection, image segmentation and images recognition. The thesis also presents following algorithms: homomorphism filtering based on wavelet transform and Ycbcrcolor space, region growing based on sub-blocks. The thesis also improved the algorithm of 8-Connected Components Labeling. The main contributions in this thesis are as follows:(1) Removing the illumination affection for images segmentation and images recognition. Because of reflect light on the body of the cow, the color in image on some part of area isdifferent from the original color. Firstly, the color space of the cow image is transformed into Ycbcr space. The wavelet decomposition is used on the illumination component, which can filter some useless details. Finally, the homomorphism filter will remove the illumination affection.(2) In order to improve the speed of recognition, the thesis presents the algorithm of region growing based on sub-blocks. Firstly, the original image is partitioned into 4×4 sub-blocks whichare not overlapped, and then the sub - blocks are segmented by color clustering algorithm. Color and texture features are extracted from the segmented sub - blocks, and a block growing algorithm is employed to achieve segmentation for the whole image based on growing rules. The algorithm of multi-threshold based on Otsu is used to separate the target from the background. Finally, the algorithm of 8-Connected components labeling is used to remove the holes and fissures on targets.(3)The black pattern with the biggest area is used as recognition target. Firstly, the target is normalized. Secondly, the cubic B-spline Function is used to construct the wavelet moment. The characteristic parameter which is generated by wavelet moment is compared by the characteristic parameter stored in Database, which can conclude the result.The experiment result shows that the algorithms which are used by the thesis can recognized cow image effectively. The recognition rate is about 84%.