Research on Improved SIFT Algorithm Based on Optimal Transmission Theory

The traditional SIFT algorithm is known for its time-consuming nature due to the high-dimensional feature descriptor and the weak anti-noise ability of matching resulting from a single matching strategy. In this paper, an improved SIFT feature matching algorithm based on optimal transmission theory is proposed. Firstly, PCA is utilized to reduce the dimension of the extracted descriptor, optimizing its structure and enhancing matching efficiency. Secondly, in order to address sensitivity to noise, a change in the matching strategy is implemented using iterative calculation of Sinkhorn distance as the distance metric and applying optimal transmission theory to establish matching relationships between descriptors. This reduces noise influence while improving stability and accuracy in image processing. Experimental results demonstrate that the proposed algorithm enhances matching accuracy across different scenarios, achieving over 90% accuracy while reducing processing time and effectively improving performance.