A New Fast Image Stitching Algorithm for Turbid Waters

A Large-Crack Image-Stitching Method with Cracks as the Regions of Interest

“…According to the method proposed by Wang et al [9], compared with using the original images, using binary images indeed reduces the number of keypoints, as shown in Figure 4a,b. In the method proposed in this study, binary images correspond to the masks of crack regions.…”

“…In the process of image stitching, the most time-consuming task is matching large numbers of keypoints. In addressing this issue, Wang et al [9] employed the Canny edgedetection method to filter out crack pixels in images and then extracted keypoints from these pixels for matching. While this approach effectively reduces the number of keypoints by reducing image information, the quality of edge detection depends on the selected parameters and may have limited effectiveness in removing noise, as shown in Fig- ure 2b.…”

“…For such cross-image crack damage analysis, image stitching can be employed to construct complete crack images for assessment [6,7]. Traditionally, the most commonly used method is to utilize the scale-invariant feature transform (SIFT) for image stitching [8][9][10]. However, while the SIFT excels in maintaining scale invariance, its drawbacks include susceptibility to the erroneous matching of keypoints, resulting in suboptimal stitching results [10]; it also involves a considerable computational burden, particularly for highresolution images, to the extent that it may be impractical [9].…”

“…Traditionally, the most commonly used method is to utilize the scale-invariant feature transform (SIFT) for image stitching [8][9][10]. However, while the SIFT excels in maintaining scale invariance, its drawbacks include susceptibility to the erroneous matching of keypoints, resulting in suboptimal stitching results [10]; it also involves a considerable computational burden, particularly for highresolution images, to the extent that it may be impractical [9]. For the issue of erroneous matching, Tian et al [10] utilized the geometric relationships between crack feature points to improve matching accuracy, while Woo et al [11] used the SIFT combined with the Random Sample Consensus (RANSAC) algorithm [12] to compute homography matrices for image stitching.…”

“…For crack image stitching, the primary ROI is the area of pixels in which a crack is located. Wang et al [9] proposed a method of first detecting cracks and then stitching images. In this study, crack detection was performed using the Canny edge-detection method.…”