A novel set of kernels for support vector machine (SVM) is presented, meanwhile, an efficient and effective feature is extracted from the images, and the al-gorithms based on it are applied on semantic image classi-fication. The experiments show that it works well on a pair of train/test set including 8000 images under the criterion of the same precision. Our algorithm can save much time in computation. Based on the experiments, the presentedapproach is 87.81% and 85.99% better than other kernels proposed for the same purpose on train set and test set respectively.
展开▼