Skip to Main Content
Semantic Image Annotation is a difficult task in Annotation Based Image Retrieval (ABIR) systems. Several techniques proposed in the past were lagging in efficiency and robustness. In this paper we are proposing a novel technique for automatically annotating multi-object images with higher accuracy. The colour entropy is used to eliminate the image background, and then we applied normalized cut principle for object separation. Our experimental results proved that the multi-class n-SVM performs better with colour feature extracted using histogram and shape feature extracted using Region Contours.