Automatic semantic annotation for image retrieval based on multiple kernel learning

Abstract

The image low level features have a gap with the high level semantic feature which human understand, the researches begin to focus on automatic semantic annotation image retrieval rather than on content image retrieval. The previous methods mostly base on single kernel learning, which has some limitations, which is no effective feature information processing. In this article, an automatic image annotation framework is proposed based on Radial Basic Kernel function combining Spatial Pyramid and Histogram Intersection Kernels. This framework utilizes multiple kernel learning, the k-mean clusters the training images to dictionary. The feature parameters are optimized Spatial Pyramid and Histogram Intersection Kernel. Then radical basic kernel function trains the data and predicts the labels of the images. Spatial Pyramid, reflecting features of location information, is more exact than Bag of Word. Experimental results demonstrated that the proposed framework effectively improves the performance of image annotation and outperform state-of-the-art on the multiple databases.