Color-texture feature extraction using soft decision from the HSV color space

Color and texture are two important low-level features of an image. These features can be used either alone or in combination using appropriate weights for content-based retrieval of images. When only color is used, retrieval results for query images with texture patterns is found to be poor. If both the features are combined, the proportionate contribution of each feature depends on the query image. This involves a high computation overhead for capturing relevance feedback from the user. We propose a new soft-decision approach to the modeling of combined human visual perception of color and texture in a single feature vector called COLTEX. It shows encouraging results in content-based image retrieval applications. We have developed a Web-based system for demonstrating our work and for performing user queries.