Analysis of Image Matching Algorithms

Image matching is a fundamental issue in computer vision. It has been widely used in tracking, camera calibration, recognition and so on. The main aim of image matching is to find the correspondence between the two images of the same scene or object in different conditions or in different environment such as difference in their viewpoints, rotations, scale, illumination, amount of blur etc,. For the process of image matching extraction of stable common features (key points) is the major issue. Many of the key point detectors can provide the information about these stable features. Scale Invariant Feature Transform (SIFT) is one of the feature point detectors that can able to provide a set of features of an image that are not affected by many of the complications experienced in other methods. As the image-matching concept involves in finding the correspondence between the two images (reference image and test image) there exists a need to find the amount of transformation involved in between the two images. For estimating the transformation, Random Sample Consensus (RANSAC) is the mostly used algorithm, RANSAC considers the stable features provided by the initial feature detector (i.e., SIFT) as an input parameters and generate amount of transformation involved in between the images in the form of a matrix. For the estimation of the transformation matrix RANSAC uses different amount of inliers for different thresholds, This transformation matrix helps us to transform the test image like the reference image so that matching accuracy increases. In this paper, we concentrate on the analysing the effect images with various environments like the view change, scale, rotation, illumination changes over the matching accuracy of the SIFT, and SIFT combined with RANSAC algorithms through experimentation.