Texture Complexity Adaptive SIFT Image Stitching Algorithm

In the field of computer vision, Scale Invariant Feature Transform algorithm is widely used in areas such as image stitching, but it has limitations when facing complex and variable image textures. In this paper, the entropy value is calculated using the GLCM to quantify the texture complexity, and a systematic improvement is made to the SIFT algorithm. In the scale construction process, the Gaussian kernel parameters are adaptively adjusted according to the texture complexity, the Gaussian kernel standard deviation is reduced in the high texture complexity region in order to acutely capture the micro texture details, and the Gaussian kernel standard deviation is increased in the low texture complexity region, highlighting the main structure of the image, realizing the optimization of the scale space, and effectively improving the accuracy and stability of feature extraction. In the feature matching stage, weights are assigned to the feature vector dimensions according to the texture complexity to inhibit mis-matching and improve accuracy, and the texture complexity-guided matching search space is constructed to narrow the search range and improve the matching efficiency. The experiments were evaluated by SSIM and PSNR.The results show that the algorithm in this paper can better fit the texture complexity changes in the spatial distribution of feature points, and the splicing quality is effectively improved, which lays a solid theoretical foundation and provides strong technical support for its in-depth application in the field of complex image analysis.