A method of image stitching with partition matching and direct detection for rotated image

Aiming at the problem of stitching overlapped images taken under different rotation angles, we present an effective and adaptive method to rotate the image to an appropriate angle before stitching in this paper. Firstly, we take the upper half, lower half, left half, and right half of the target image as the four partitions to be matched. Then, we make feature extractions and match between the reference image and those four partitions of the target image respectively. According to the half part with the most matching pairs, the target image is rotated by a preliminary angle. At the same time, the distribution area of the matching points of the reference image is calculated as the overlapping area. After matching the overlapping areas of the reference image and the rotated target image, we propose a method to detect the direction of the matching pairs to adjust the rotation angle. Finally, the reference image and the rotated target image can be successfully stitched into a wider image. Experimental results show that our method can find a suitable angle to rotate the target image, and can achieve an effective and satisfactory stitching result.

Automated summary

This paper presents an effective and adaptive method for rotating images to an appropriate angle before stitching to solve the problem of stitching overlapped images taken under different rotation angles. The method involves partitioning the target image into four parts and extracting features to match them with a reference image. The rotation angle is adjusted based on the part with the most matching pairs, and the direction of matching pairs is detected to further refine the rotation angle. Experimental results demonstrate the effectiveness and satisfactory stitching result of the proposed method.