Visual Interest point detection based Rank Order Log filter for Content Based Image Retrieval

With increased digital images in image repositories efficient indexing and searching becomes essential for large image archives. Content based image retrieval system (CBIR) retrieve relevant images using low-level features like color, texture and shape for image representation. In this approach, a scheme of novel color image retrieval based on visual interest point is presented. The existing systems use detectors such as Harris Laplace transform, LOG filter and Harris Detector for the extraction of visual interest points. Such detectors face drawbacks such as the scaling, distortions which affects output image, by the detection of multiple peaks around the corner or blob and leads to the production of multiple false peaks. The proposed approach using Rank Order Log Filter is more robust to abrupt variation of images caused by both illumination and geometric changes, detecting multiple false peaks around the blob and avoids them enabling the extraction of visual interest points.TheFirst step involves conversion of a query image into 8 bitplane images. Higher bit plane reflects the main content of the primitive image and it is the significant bitplane.An improved Rank Order LOG filter is proposed for the extraction of visual interestpoints.The visual interest points can be quantized and color space transformation is done in future for mapping the extracted visual interest points to color histogram for similarity measure thus retrieving the relavant images. IndexTerms—Bit plane images, Improved Rank Order Filter, Content based image retrieval.