Influence of Different Descriptors to Enhance Image Registration Techniques Using FREAK: Case Study

Image registration is considered as an important research direction in image processing and computer vision. Image registration is the method of arranging, matching and overlaying two or more images of a scene which are captured from similar scenes, but not same scenes. Images captured at different times from different viewpoint may vary in contrast, color or brightness. Image registration transfers the color style of target images to the reference image selected from one of these captured images. This paper evaluates the performance of different descriptors like SIFT, SURF, FREAK using SIFT, FREAK using SURF and CNN in generating matching images. Different performance measures like SSIM, MSSSIM, CSSS, MSE, PSNR, UQI and RMSE are used to compare the matching image with the input reference image to determine the visual quality and structural similarity. Experimental results show that FREAK using SURF outperforms other descriptors in the case of structural similarity and visual quality.