A Novel Signature Histogram Based Indexing For CBIR System

Abstract: Content Based Image Retrieval (CBIR) is a technique used for retrieving similar images for given input image from an image database. CBIR aims to retrieve the images based on the content of a given image rather than textual information of a file name. CBIR uses the various features such as color, texture, shape etc.The shape is independent of transformations like scaling, translation, rotation and flip. A good shape representation method retrieve similar images irrespective of the transformation performed on a shape. The known shape representation called “centroid contour distance (CCD) signature” is invariant to translation, scale but not to rotation and flip. The CCD signature quantize into signature histogram which is invariance to rotation and flip. It has a drawback namely number of false positives will increase i.e. many different shapes can have a “similar signature histogram”. This problem can be solved with augmented signature histogram with information present at the boundary of object. Images are presented to a user calculating Euclidian distance matching with every shape descriptor is a computational expensive problem. With large database this process may take minutes, hours. So to reduce the computational requirement, the images are indexed based on Color, Medial Axis, Area, Eccentricity, and Euler number.These indexing parameters are bias. To overcome this problem the static parameter such as mean was calculated from the signature histogram. Based on mean value images are indexed. This reduces the search space in retrieval process. So that the performance is increased.Keywords: Signature histogram, centroid distance signature, shape signature, indexing.

Keywords: Signature histogram, centroid distance signature, shape signature, indexing.