Ridge Matching Based on Maximal Correlation in Transform Space

Abstract

Image matching, a common technique in Computer Vision to identify objects, persons, locations, etc., is widely used in both military and civilian applications. For common image matching algorithms, results may vary when the raw images are captured under different lighting conditions. To reduce the unwanted influence from ambient lighting, we propose a novel method to match images that contain ridge features. The new method uses an established ridge detection algorithm to reduce raw images to sets of ridge points, each point defined by its orientation and location. To perform ridge matching, we find the pair-wise transform between every ridge point from one image and every ridge point from another. The result is a point cloud in transform space. The correlation between two sets of ridge point is equivalent to the density of the point cloud, computed by convolving the point cloud with a blurring kernel. The best match is found as the location in transform space at which the correlation reaches global maximum. We tested the new method on two image pairs; the first image pair contained artificial ridge features and the second pair was sampled from a high resolution image of the human palm. Both tests returned accurate results.