Segmented Image Retrieval Using MPEG-7 Similarity

Content-based image retrieval searching for visually identical or nearly-identical images to a query image utilizes global features of images to measure the visual similarity of images. In this paper, we propose a method to retrieve images with visual similarity measured on segmented images based on MPEG-7 visual features and an image segmentation method that utilizes the MPEG-7 visual features as homogeneity criteria. The experiments demonstrate that the image retrieval can benefit from the segmented images with homogeneity criteria based on visual features normally used for image retrieval. The experimental results show that the segmented image retrieval outperforms the conventional image retrieval.