Matching method of partial shoeprint images based on PCA-SIFT Algorithm
Abstract
To improve the accuracy of image matching shoeprint image feature matching method based on PCA-SIFT is proposed. Firstly, feature detection and pre-matching of images are done by using PCA-SIFT (principal component analysisscale invariant feature transform) algorithm. And then, the correlation coefficient is used as similarity measurement, which can filter image interest points. By this method, the image matching pairs can be obtained. Finally, the RANSAC (random sample consensus) algorithm is used to eliminate the mismatching pairs. The simulation results demonstrate that the proposed algorithm is more robust while maintaining good registration accuracy when analyzing partial shoeprint images in the presence of geometric distortions such as scale and rotation distortions compared with conventional algorithms.
Keywords
Download Options
Introduction
As a form of physical evidence, a shoemark can provide an important link between the criminals and the place where the crime occurred. It has been reported that there should be equal and perhaps even greater chance that footwear impressions could be present at a crime scene, compared with the presence of latent fingerprints [1,2,3]. Nonetheless, footwear impressions have great potential in assisting forensic investigations. For instance, for a repeat offender who may commit a series of offences in a relatively short period of time, it would be unusual to discard or change his/her footwear between different crime places [4,5].
In this paper, the issue of automatically classifying shoemarks is addressed. A critical issue that has to be overcome in order to achieve such a goal is the fact that one may have no control over the quality of the shoemarks collected from Scene Of Crime Officers ( SoCs)[6]. As partial shoeprints, resulting from external destruction or incomplete contact between shoe sole and the ground surface are commonly found in crime scenes, the performance of a retrieval system for partial prints is of considerable interest and importance. As shoeprints left in crime scenes may be incomplete impressions, this makes many shape descriptors unsuitable for the application. One of the potential solutions to the problem is extracting features of local interest points because such features can still express the pattern even though it is a partial shoeprint[7,8].
In this study, a new matching method is proposed based on PCA-SIFT algorithm [9,10]. Firstly, feature detection and prematching of images are done by using PCA-SIFT algorithm. Then we applied the matching between the extracting interest points descriptor with a nearest neighbor method using the Euclidean distance. Secondly, the mismatching is wiped out by using RANSAC algorithm [11,12]. This method solves the mismatching problem of image matching.
Conclusion
This work proposed local invariant features and key point matching for recognition and retrieval of shoeprint images. The experimental results also show that the performance of matching toe prints and heel prints as well as that of matching lefthalf prints and right-half prints. Correlation coefficient, as one of the similarity measurements, can effectively measure the degree of similarity between two samples. In this paper, we combine the PCA-SIFT algorithm with the correlation coefficient to get preliminary match pairs of two images. To improve accuracy of matching, the RANSAC algorithm is utilized to delete false matching points by noise and duplicate pairs. Finally, we can get one-to-one correct matching point pairs. In order to demonstrate the superiority of this method, we show the simulation results. Experimental results show that compared with conventional algorithms, the proposed algorithm is more robust while maintaining good registration accuracy when analyzing partial shoeprint images in Scene of Crime Officers.