An image classification approach based on sparse coding and multiple kernel learning

A novel image classification method based on sparse coding and multiple kernel learning is proposed in the paper. Traditional methods of image classification used common sparse coding but lose the spatial information. We add this spatial information by dividing the image with the spatial pyramid. With the nonlinear SVM for image classification, each level of spatial pyramid has its own kernel, and we adopt machine learning for the optimal trade-off between different kernels. A much more discriminative kernel can be seen as the linear combination of base kernels corresponding to different pyramid levels. The experiments on the benchmark dataset show the effectiveness and robustness of our method. The precision on scene categories dataset can reach 83.10%, and it is the best result comparing to the state-of-the-art work.