|
|
An Improved Image Registration Algorithm Based on SURF⁃ORB |
SHANG Mingshu1, WANG Kechao1, GAO Yubao2 |
1.School of Information Engineering, Harbin Institute, Harbin, Heilongjiang 150080, China
2.Exhibition Hall of Criminal Evidence of the 731st Army of the Japanese Invading China, Harbin, Heilongjiang 150001, China |
|
|
Abstract An improved image registration algorithm based on SURF-ORB is proposed. Establish speeded up robust features (SURF) image pyramid, detect oriented FAST and rotated BRIEF(ORB)feature points on it and use 256 bit binary characters as feature vectors to describe feature points. The nearest neighbor method matches feature points. Filter matching points by utilizing the similar properties of neighborhood average gray difference, Euclidean distance, matching point line and x-axis angle among the correct matching points. The k-means clustering (k-means) algorithm is improved, using points with a density greater than the threshold as the center point of the class, clustering, deleting classes with a sum of squared errors greater than the threshold and reclassifying the remaining feature points into the reserved classes. The random sample consensus(RANSAC)algorithm is improved, merging all transformation matrixes’ interior points into a set and classifying the matching points in the set that meet the error distance threshold with the candidate optimal transformation model into its inner points. Use the least squares method to recalculate the transformation matrix with all its interior points to obtain a more accurate solution. The experimental results show that the number of feature points extracted by this algorithm is about 32% less than that of SURF and ORB algorithm, the matching accuracy is improved by about 16%,and the operation time is increased by about 0.26%.
|
Received: 28 May 2024
Published: 19 March 2025
|
|
|
|
|
|
|
|