Multi-Task Joint Spatial Pyramid Matching with Dynamic Coefficients for image classification

recognition considered as a necessary part in many computer vision applications. Recently, Sparse coding methods which based on representing a sparse feature from image, show remarkable results on several image recognition benchmarks. However, the precision obtained by these methods are not sufficient. Such a problem arises where there are a few number of training images available. So using multiple features and multi-task dictionaries seems to be essential to attain better results. In this paper, we use multi-task joint sparse representation but unlike previous works [4] which used constant coefficients for all class, we apply dynamic coefficients which computed separately for each class to combine these sparse features efficiently. In other word, we calculate the importance of each feature for each class separately. It causes the features to be used more effectively and better appropriate for each class. We used PSO (Particle swarm optimization) algorithm to obtain these dynamic coefficients. To the best of our knowledge, our experimental results on Caltech-101 and Caltech-256 databases surpass in accuracy from the best published results to date on the same databases