Incorporate discriminant analysis with EM algorithm in image retrieval
Qi Chuan Tian, Ying Wu, Thomas S. Huang · 2002
One of the difficulties of content-based image retrieval (CBIR) is the gap between high-level concepts and low-level image features, e.g., color and texture. Relevance feedback was proposed (Rui et al., 1999 to take into account the above characteristics in CBIR. Although relevance feedback incrementally supplies more information for fine retrieval, two challenges exist: the labeled images from the relevance feedback are still very limited compared to the large unlabeled images in the image database; and relevance feedback does not offer a specific technique to automatically weight the low-level feature. In this paper, image retrieval is formulated as a transductive learning problem by combining unlabeled images in supervised learning to achieve better classification. Experimental results show that the proposed approach has a satisfactory performance for image retrieval applications.