Comparison Between the Traditional and Deep Learning Algorithms on Image Matching

Image matching is the cornerstone of many computer vision tasks, which has high requirements on matching accuracy and time-consuming. The steps of extracting features, calculating descriptors, brute force matching, and filtering outliers have been integrated into a complete pipeline. This paper compares the typical traditional algorithms SIFT and RANSAC in the field of image matching with the new mainstream algorithms SuperGlue and AdaLAM proposed under the background of deep learning in recent years through experiments. Experimental results show that although traditional algorithms are still highly competitive, they are far less effective than deep learning algorithms for images with poor texture.