Improvement of multi-scale image alignment techniques in computer vision based on numerical analysis methods

Image alignment is a fundamental problem in the field of computer vision and an important prerequisite for carrying out many other tasks.Firstly, the theoretical basis and realization method of image alignment as well as the process and the method of alignment are introduced to provide alignment ideas.Subsequently, an image alignment method based on the union of multi-scale features is proposed, and a new loss term is introduced to the small-scale features therein, which further improves the distinguishability of the small-scale feature descriptors while guaranteeing the invariance of the large-scale feature descriptor matching therein.Three common alignment algorithms (RIFT algorithm, HAPCG algorithm, and SAR-SIFT algorithm) are selected for stability assessment and quantitative evaluation on the dataset, and an image enhancement algorithm with histogram equalization is used to enhance the dataset.The results show that the feature stability of this paper's method is described as 99.1%, which is better than other algorithms.Meanwhile the desired effect is achieved on the dataset.