Image Stitching Algorithm Based on the Longest Common Subsequence

图像配准和拼接技术在医学、军事、遥感、航空、农业等众多领域有着广泛应用，是模式识别、虚拟现实的关键技术之一，针对现有方法在亮度差异大、纹理稀疏场景中配准失败率高的问题，本文提出了一种图像拼接算法，通过计算图像灰度值差序列的最长公共子序列来确定重叠区域，并自动定位拼接线。该算法提出多尺度融合策略优化拼接边界，引入灰度差分序列替代原始像素，消除系统性亮度差异，无需手工指定特征点，对亮度差异不大的图像拼接具有较高的精度和效率，适用于多种图像拼接情形，包括同高度、不同高度、有平移和亮度差异等情况。实验结果表明，该算法能够实现快速、无缝的图像拼接，在亮度突变、噪声干扰(SNR ≤ 15 dB)等复杂场景下，本算法的拼接成功率较SIFT提升23.6%，耗时降低48.3%，可处理±30˚旋转和1.5倍尺度变化，具有良好的鲁棒性和实用性。Image registration and stitching technology is widely used in many fields such as medicine, military, remote sensing, aviation, agriculture, etc. It is one of the key technologies of pattern recognition and virtual reality. In response to the existing methods’ problems of high registration failure rate in scenes with large brightness differences and sparse textures, this paper proposes an image stitching algorithm, which determines the overlapping area by calculating the longest common subsequence of the image grayscale difference sequence and automatically locates the stitching line. This algorithm proposes a multi-scale fusion strategy to optimize the stitching boundary, introduces a grayscale differential sequence to replace the original pixels, eliminates systematic brightness differences, and does not require manual specification of feature points. It has high accuracy and efficiency for stitching images with little brightness differences. It is suitable for a variety of image stitching situations, including the same height, different heights, translation and brightness differences. Experimental results show that this algorithm can achieve fast and seamless image stitching. In complex scenarios such as brightness sudden change and noise interference (SNR ≤ 15 dB), the splicing success rate of this algorithm is increased by 23.6% compared with SIFT, and the time consumption is reduced by 48.3%. It can handle ±30˚ rotation and 1.5 times scale changes, which is good robustness and practicality.