Visual Keyword Based Image Retrieval

This paper proposes the visual keywords to represent the visual appearance, such as color and texture, of regions in images without precisely image segmentation. Based on the proposed visual keywords, a multi-modal image query and retrieval (MIQR) approach is presented to retrieve desired images from image databases by using textual annotations associated with images and/or the proposed visual keyword. In addition, the Generalized Probabilistic Decision-Based Neural Networks is adopted to model the visual keywords. The experiments were performed on a subset of the COREL image gallery and the images gathered from Internet. A comparison with current leading approaches is made. The experimental results show that the MIQR approach can retrieve relevant images closely associated with users' query regions or objects, and has the capability of searching and retrieving relevant images from a large collection of images.