Face recognition based dog breed classification using coarse-to-fine concept and PCA

Dog breed classification or identification is important for dog training, and curing. The conventional identification method is based on experts which is hard to find. This paper proposes a method to classify dog breed based on the dog face images. The proposed method is based on the coarse to fine concept, where the template matching technique is applied for coarsely classifying the images into 5 groups. Then, within each group, the principle component analysis (PCA) is applied to finely classifying the dog breed. In the PCA- based classification, face features are represented in term of a weight vector. A set of sample image of each dog breed are used for learning the features of the breed. The average weight vectors are stored as the templates of features for breeds after the coarse classification. During the running time after the coarse classification, a dog face image is passed through the PCA to find it vector representation. This vector will then be compared with feature template for each breed in the database. The image under test is classified as the breed that gives the minimum distance between the twp vectors. To evaluate the performance of the proposed method, experiments with 700 dog face images from 35 dog breeds had been performed. Before the testing, 5 dog face images from every breed (totally 175 dog faces) were used to train the system. The experiments showed that the proposed method (coarse classification and PCA for fine classification) gives approximately 93% accuracy which is better than the PCA-based classifier without the coarse classification. The improvement is 16% approximately.