Feature based multi-view image registration using SURF

This paper presents an efficient method for multi-view image registration based on SURF (speeded up robust feature) to enlarge a field of view. Multi-view image registration is used in the field of video conferencing, 3D image reconstruction, generating large field of view and satellite imaging. It can also be used for mosaicing based localization, shape recovery and motion detection and tracking. The main contribution of the proposed work resides in the primary detection of the features using SURF feature detector and descriptor. This step play a vital role as it detects the robust features and describe the detected feature for matching purpose. These features are matched using SSD (Sum of Square Difference). Then remove the false matched pairs for finding the accurate homography matrix using the RANSAC (Random Sample Consensus) algorithm. The final mosaic image is obtained by warping the images into a single plane, and applies the multi-band blending which make smooth transition between the images of the same scene. Also, comparison of the SURF and Harris are shown as a base of the feature detection step. The experimental result shows that, the present method using SURF is robust, which detect the features with higher precision and get a perfect panorama.