为了对身管内壁序列图像进行精确配准与融合拼接, 得到大视场高分辨率待检测图像, 根据图像特点提出了一种结合边缘检测的快速SIFT图像拼接方法。该方法充分考虑待处理图像的特点, 首先对感兴趣区域的图像进行边缘检测, 分割出细节信息最丰富的子区域, 再对分割出的子区域提取SIFT特征点并进行配准。然后, 使用基于Sigmoid型函数权重的图像融合方法, 实现图像之间的无缝融合, 最大程度地保证了融合图像的清晰度和细节信息的完整性。实验结果表明: 改进的方法和传统SIFT算法相比, 在特征点提取阶段的平均效率提高了80%左右, 且整体配准阶段的效率也有较大提高。图像融合结果在主观评价和各种客观评价值上都能满足工程实际需求。

In order to accurately register and stitch the sequence images of the inner wall of the barrel, and get a image with high field of view and high resolution, a fast SIFT image stitching method with edge detection according to the characteristics of the overlap region of the images was proposed. It took full account of the characteristics of the images and could quickly segment the sub-region that possessed the most abundant anomalous information by detecting the edge of the region of interest. Then, it extracted SIFT feature points of the sub-region and matched them accurately by RANSAC. After that, a novel fusion method based on the weight of Sigmoid function weight was used to realize the seamless fusion between sequence images. This method can maximize the clarity of the fused image and the integrity of the detailed information. Experimental results show that the improved algorithm is much less time-consuming than that of traditional SIFT algorithm. Its computational efficiency has improved about 80% in feature points extraction process and the efficiency of the whole registration process has been also improved. The subjective evaluation and the various objective evaluation values of the fusion results by this fusion method are superior to other fusion methods.