Application of Improved KAZE Algorithm in Image Feature Extraction and Matching

Intelligent navigation and recognition technology have continuously improved the field of image matching, so how to achieve more efficient and accurate feature matching is the key to image processing. This study provides a detailed introduction to the calculation methods, advantages and disadvantages of various algorithms such as SIFT, ORB and KAZE. Finally, the KAZE algorithm was selected because it can better handle edge information. However, this algorithm still has drawbacks such as high computational cost. Therefore, starting from the feature description and similarity calculation, the study introduced the concepts of second-order degree values, circular rotation characteristics, and expected distance and intercept distance to improve it, which is called the algorithm. Theoretically, this algorithm can greatly reduce the computational burden of the algorithm. To further validate the effectiveness of the improved algorithm, the study selected five types of image groups from the Mikolajczyk standard image database, including changes in image scale, brightness, and perspective. Simulation experiments were conducted on the improved algorithm and the remaining five algorithms. The experimental results show that compared to the original algorithm, Li-KAZE reduces the overall running time by 27.9% and improves the average matching rate by 0.12%. Compared to other SIFT and other algorithms, Li-KAZE also has the best performance among all algorithms. When faced with changes in compression values and extensive changes in graph groups, its highest repeatability reaches 99%. In summary, the comprehensive performance of the Li-KAZE algorithm has achieved the goal of improvement.