改进最佳缝合线的红外图像拼接方法

变电站场景的红外图像受噪声干扰大、纹理信息不明显，拼接过程中易出现拼接痕迹或重影现象。针对上述问题，本文提出一种改进最佳缝合线的红外图像拼接方法。该方法首先采用SIFT算法提取图像区域特征，实现图像配准；然后在两幅图像的重合区域上引入局部权重系数，并对图像颜色差异强度进行形态学操作，减少红外图像的噪声干扰，以此改善能量函数图的纹理信息。最后，利用动态规划改进缝合线搜索准则，在图像重叠区域搜索出最佳缝合线。实验结果表明，与渐入渐出法和最佳缝合线法比较，本文方法在拼接后图像的平均梯度、图像清晰度和图像边缘强度均有所提高，融合区域过渡更平滑自然，拼接痕迹明显减少。

Infrared Image Mosaic Method for Improving the Best Seam-line

The infrared image of a substation is significantly disturbed by noise and the texture information is unclear. Therefore, stitching traces or the ghosting phenomenon may appear in the process of stitching. To overcome these challenges, this study proposes an infrared image splicing method that improves the best seam-line. First, this method uses the SIFT algorithm to extract the image area features to achieve image registration, then introduces local weight coefficients in the overlapping area of the two images. Subsequently, morphological operations are performed on the intensity of the image color difference, which reduces the noise interference of the infrared image and improves the texture information of the energy function graph. Finally, dynamic programming is used to improve the seam-line search criteria and search for the best seam-line in the image overlapping area. The experimental results show that compared with the gradual fusion method and the best seam-line method, the average gradient, image clarity and image edge strength of the stitched image are improved, the transition of the fusion region is smoother and more natural, and the stitching trace is significantly reduced.