Research On Image Feature Matching Based On The Fractional Calculus Approach

The main purpose of digital image processing is to improve image quality,image features and etc,for identifying and analyzing by human or computer. With the development of the society modernization, high quality of science and technology is being expected, image processing has already been applied to almost all scientific research, engineering and human life. And more and more high precisions are required. Image feature extraction is a crucial step in all image analysis tasks in image texture description, image matching, image segmentation and other image processing applications.With development of the fractional calculus, especially when it is observed that the description of some systems is more accurate and effective with the fractional derivative theory, the fractional calculus theory has being found an increasingly wide utilization in all fields. Compared with the integer calculus, fractional calculus has greater advantages for signal processing. Not only can fractional differential operation effectively enhance the image, but also can retain image texture detail of high quality. This is the fundamental distinction between integer calculus and fractional calculus. Consequently, it is a kind of fresh thinking and direction to introduce the fractional calculus into image processing.In this paper, by using theoretical results of the fractional calculus in the image feature extraction and image enhancement, combining with feature-based matching algorithm, mainly SIFT feature matching algorithm, improved image matching algorithm is researched and discussed.SIFT (Scale Invariant Feature Transform) image matching algorithm is mainly based on local features. Therefore, the stability of features of image is critical in image matching. In this paper, fractional calculus theory is introduced into image matching, to improve the quality of image features. Based on theoretical study and experimental of SIFT, An improved SIFT integrating Gaussian filter with fractional calculus filter is presented in this paper. Compared with the classic SIFT, the results of many experiments certify that the SIFT image feature matching based on fractional calculus not only extracts more keypoints, but also can improve the stability of SIFT feature keypoints. Generally speaking, the improved SIFT algorithm can improve the accuracy of matching and robustness of the algorithm. It also can reduce mismatches. Especially in the rich texture image matching, the effective of matching is better. To sum up, the improved SIFT algorithm outperforms the original one.