Feature Detection Based On FAST Detection And SIFT Description

Recently, with the rapid development of computer vision technology, especially the abundance of the popularity of smart phones, the number of images grows rapidly due to the convenient method of user obtains it. And digital image contains more information compared with other forms of documentation, so it becomes the most important information transmission way. But, when we face with a large amount of images data, select reasonable detection and description considering various of angle, illumination, scale and other factors, which able to reduce the interference of background clutter and noise. To realize the image features identify reliable and classify accurate, it became an important research topic. The main content of this paper is to develop an algorithm would detect and identify the specific target object in the scene. The details are as follows:First, the local feature are detected from particular scene by feature detection algorithm. This paper introduce several traditional local features detection algorithms, and compare the feature point numbers and detecting speed. We concluded that FAST extract a lot of feature points in short time.Then, apply SIFT features described algorithm which provide high-dimensional scale invariant characterization. In the feature matching stage, we define the SNR of the feature points, and recognize the object in the noisy background adopt high robustness Bayesian classifier to recognize the object in the noisy background.Finally, the experimental result shows that the algorithm of FAST combine with SIFT has both high efficiency of FAST and high accuracy of SIFT description.