General Image Retrieval Using Shape and Combined Features

Content-based image retrieval (CBIR) is an important multimedia application. In recently emerged MPEG-7 standard, image content is described by perceptual features such as color, texture and shape. Many shapebased image retrieval techniques have been proposed in literature. However, almost all of these techniques assume isolated objects and use only shape feature for retrieval. General image retrieval or retrieval based on segmented objects are little addressed. In this paper, a general image retrieval technique is presented to address image retrieval based on segmented objects. The objects in the image are extracted through a semi-automatic segmentation tool, and are indexed using both individual features and combined image features. A Java-based client-server retrieval framework has been developed to conduct the online retrieval experiment. Results show that general image retrieval using shape features performs well for images with rigid objects, and image retrieval using combined features outperforms retrieval using individual features.