基于Harris特征和结构相似约束的图像拼接方法研究-计算机技术专业论文.docxVIP

万方数据 万方数据 ABSTRACT Image mosaic is a hotspot in image processing which has been successfully applied in remote sensing, medical image processing, video surveillance, virtual reality, military and other fields. Meanwhile, the requirement of these applications reacts on the image mosaic and promotes its development. The demand of the image mosaic applications for accuracy, speed and robustness is becoming much higher. This paper carried intensive research on Harris descriptor and similar structure constraint, and put forward an efficient mosaic algorithm. The main contribution of this paper are summarized as follows: In the feature extraction and description, the classical Harris corner detector which is invariant to rotation, brightness and contrast changes has been studied. To solve the problem that the Harris corner detector is not invariant to image scale, the multi-scale Harris corner detection based on B-spline was proposed. Single-scale Harris corner detector is transformed to multi-scale by B-spline. Keypoints were described by BRISK descriptor which is a robust and fast binary bit string. In the feature matching, the traditional RANSAC（Random Sample Consensus） was improved by using Hamming distance as similarity measure, and kd-tree as the search strategy for the initial matching. Similar structure constraint was used to purify initial matching. Then, RANSAC was utilized to calculate the transformation matrix. This method has eliminated the false matching points and reduced the iteratives of RANSAC, thus improves the precision and speed of matching. In terms of image fusion, a method was given for calculating the overlapping area, and using the weighted method to fuse overlapping area. Due to the color differences caused by different cameras, or images captured by the same camera but different times, the color transfer algorithm was adopted to adjust the color after fusion, to produce consistent tone of image. V Key Words: image mosaic, B-spline, Ha