An Applicability Analysis To SIFT And Harris Corner Detector For Feature Points Extraction

In recent years, with the rapid development of digital image processing technology, image feature matching technology has been widely used in many fields. The primary task of image feature matching is extracting the stable image feature points from the matching image, and then using relevant feature points for establishing the corresponding relationship between the original image and the matching image. Selecting an appropriate point feature extraction method is vital of extracting feature points, in view of the point feature extraction methods have an influence on the speed and precision of image matching, so selecting an appropriate point feature extraction method has certain theoretical significance and practical value.The paper focuses on image feature points extraction: first of all, to select values for the noise reduction threshold of sift algorithm, then we have carried on the analysis of the applicability of the Harris corner detection algorithm and sift algorithm in image feature points extraction. The specific research contents are following:(1)To avoid the blindness of selecting noise reduction threshold in SIFT algorithm extracting image feature points, as well as inappropriate noise reduction threshold has a bad influence on SIFT algorithm, the purpose of this paper is to study whether there is a stable noise reduction threshold during extracting image feature points in SIFT algorithm, in order to obtain a noise reduction threshold which can lead to the best results. Based on multi-groups of the same restrictions of experimental images(including the original image, the rotation image and the noising image), to select different noise reduction threshold for experimenting. The experimental results show that when the noise reduction threshold is approximately 30, this can let the SIFT algorithm obtain relatively stable feature points.(2) We respectively analyze the process of extracting feature points of Harris corner detection algorithm and SIFT algorithm. As the SIFT algorithm is more complicated, meanwhile the process is more complex to build the Gauss difference scale space, considering the process of Harris corner detection algorithm is relatively simple. In order to intuitively compare the two kinds of algorithms, here we use the three aspects(feature point availability and computational efficiency and feature point similarity invariance) which are mentioned in this paper. Meanwhile in the condition of the same experiment images, We using them for quantitatively studying the two algorithms.(3) In order to explore the applicability of Harris corner detection algorithm and SIFT algorithm in the feature points extraction of different kinds of images. Here for two kinds of images: polyline feature dominant images and smooth curved line feature dominant images; three aspects: feature point availability and computational efficiency and feature point similarity invariance; three indexes: the image matching ratio, sample mean and standard deviation. We use them for quantitatively studying the two algorithms. Finally, the experimental results fully show that for polyline feature dominant images, Harris corner detector presents better performance than SIFT in all three aspects. However, for smooth curved line feature dominant images, as a result of too few feature points are detected, the applicability of Harris corner detection algorithm is not as good as SIFT algorithm.In the end, we summarize all the relevant work of this paper, Moreover, we design a prospective about the research content, combining with the shortcomings of this paper.