Research On Image Mosaic Based On BRISK Algorithm

There are two approaches generally used to obtain wide angle of view images for the large object scenes in the computer vision community.One method is shortening the camera’s focal length to shoot the complete scene, but it will reduce spatial resolution of images.The other method is useing fisheye lens, wide-angle lens and reflection/refraction optical system, this method need complex and expensive hardware devices, and easily produce larger distortion near the edge of an image.With the development of computer technique, the technology of image mosaic offers a new idea to solve these problems. Image mosaic is to use the image automatic registration techniques to align two or more images with overlapping area, then use the image fusion technology to generate an equivalent image with wide viewing angle and high image resolution. This method decreases the hardware requirement of cameras, narrows the equipment, and has the advantages of intelligence, automation, and low cost. Therefore, the algorithms of image mosaic are studied in this article.This thesis investigated the developing status and trends of image mosaic technology, and the theory of image mosaic is generalized and summarized. The key techniques, such as the image preprocessing methods, the image registration algorithms, and the image fusion technology are analyzed and expounded. The classic image feature matching algorithms, such as Harris, SIFT, SURF, BRISK, FAST, and AGAST, are analyzed in-depth, and this laid the foundation for further work.To get accurate space stitching images, this paper proposes an algorithm combined with BRISK and cross-correlation template matching algorithm. There are some considerable errors in the splicing translational direction, when using traditional BRISK algorithm. To counter this problem, BRISK algorithm can be used for the image of the scale and rotation correction.Then the template matching method can be used to solve the translation correction.At the same time, RANSANC algorithm is added to the BRISK algorithm to achieve precise stitching.Some of the nomonymy points are picked out as cross monitoring points in the overlapping region.Then calculating the root- mean-square error of the distance between pixel coordinates of corresponding monitoring sites, and the image matching precision can be quantified.The experimental results with database images show that it is a kind of short operation time, high precision image matching method with good results.