A Novel Generalized SVM Algorithm with Application to Region-Based Image Retrieval

Support vector machines (SVM) has been widely applied in the area of content-based image retrieval in order to learn high-level concepts from low-level image features. Most existing SVM based image retrieval algorithms only rely on global-based features to represent the image content, which obviously can not well reflect the image semantic content. Region-based representations are far more close to the image content. However, such representations are of variable length and the Gaussian kernel is inappropriate in this situation. In this paper, a novel generalized SVM algorithm is proposed, which takes into account both low-level features and structural information of the image, in order to solve the problem of region-based image retrieval via SVM framework. Firstly, for a given image, salient regions are extracted and the concept of salient region adjacency graph is proposed to represent the image semantics. Secondly, based on the SRAG, a novel generalized structure kernel based SVM algorithm is constructed for content-based image retrieval. Experiments show that the proposed method shows better performance in image semantic retrieval than traditional method.