Evaluation of distinctive color features from harris corner key points

In image retrieval and image registration scenarios, interest point detection is essential to find regions in which descriptors are calculated. Most of the current methods use only the intensity information of the images to find the key points. We investigate the use of color information in interest point detection. Corner points are extracted from images using Harris corner detector on gray image (i.e. using gray level intensities) and on color image (i.e. using color information in RGB space). The extracted corners are analyzed for application to image matching using cross correlation and Random Sample Consensus (RANSAC). Cross correlation is used to find the matches, and RANSAC to reject inconsistent matches and find the inliers, which are considered as correct matches. The performance of the feature detection methods are compared for view point changes, rotation, blur, illumination changes. All the experiments use repeatability measurement and the number of correct matches for the evaluation.