An improved image mosaic based on Canny edge and an 18-dimensional descriptor

• DOI: 10.1016/j.ijleo.2014.04.069
• Publication Date: 2014-07-10T20:02:03Z
• Authors: Yong Chen; Min Xu; Huan-lin Liu; Wei-na Huang; Jiang Xing
• Keywords: Scale-invariant feature transform; RANSAC; Canny edge detector; Image gradient; Feature (linguistics)

Abstract To gain the wide view angle and high resolution image stitched by the sequence images overlapped in the same scene, this paper proposes a new image mosaic method of combining the improved SIFT algorithm with Canny feature edge detection based on the traditional scale invariant feature transform (SIFT) algorithm. First, we extract feature points of the image with the SIFT algorithm. Then, we construct an improved 18-dimensional feature descriptor consisting of 12 gradient values in a circular window, three cumulative gray-scale values and three gray-scale differences in a concentric circular window. Second, this method preserves the features in 16-neighborhood of the Canny edge image. Then, the extracted feature points search for the coarse matching points through the BBF, and the coarse matching points are purified with the random consistency (RANSAC) algorithm, which calculates the transformational matrix H among the images with these purified feature points. Finally, the seamless mosaic of images is completed with the image fusion algorithm of slipping in and out. In the experiment, we process the aerial image and the images that had changed in terms of translation, rotation and zoom, and parallax with the traditional SIFT and the proposed method, respectively. The results indicate that mosaic time, respectively reduced by 11.8%, 13.5%, 7.7% and 21.6%, the root mean squared error, respectively reduced by 80.1%, 0.002%, 4.4% and 58.1% through the proposed method in this paper, compared with the SIFT. Even for an image subjected to such situations as illumination, rotation, scaling and affine distortion, the matching performance can be improved with the proposed method.