Visual and textual information fusion using Kernel method for content based image retrieval

In computer vision, each region of an image has an equal and important value that is either an object in an image or the text. The appearance of text within images is certainly a rich information for humans as well as for machines. So far, the art methods have explored either visual features or the social tags to retrieve similar images. Another possibility for image retrieval is to generate fully automatic tags/keywords by extracting embedded and scene text in images along with low-level visual features and fuse them together. Considering this, we have investigated a novel approach to retrieve similar textual images by exploiting visual and textual characteristics of the image. The method extracts visual salient features in first step, and the text is detected and recognized in next step. The method allocates two feature vectors each for visual and textual, and fused them together using Kernel method. The method supports three modes of search: Image query, Keywords, and Combination of both. The experimental results on benchmark datasets shows the textual features can be as effective as visual features for CBIR applications.