The Shoeprints Classification Identification Method Based On Frequency Domain Characteristic

The shoeprint is a kind of trace evidence which is often found at crime scenes. The trace evidence can help police to ascertain merits and analyze crimes. It also plays an important role in joint cases and provides valuable forensic evidences. The effective management of shoeprints can shorten the handling time of the police department and improve the efficiency of the criminal departments. At present, the shoeprint arrangement methods of public security organizations are different, most of them use manual operation and manual management method, the form of the method is single, which causes shoeprint can not play a role like the fingerprint in the process of criminal case detection. In this context, it is necessary to find a kind of image processing algorithm based on pattern recognition, which takes full advantage of the rapid and efficient characteristics of computer to realize the automatic shoeprint classification and recognition, and then make the shoeprint play a better role in criminal cases detection.By observing a large number of the actual shoeprint images, we found that many shoeprint images contain equally spaced strips pattern, concentric circles pattern and other periodically repeated pattern. According to Fourier transform characteristics, frequency spectrum of periodically repeated signal (pattern) in time domain (spatial domain) have apparent peak points. The research of this thesis is focused on the shoeprints classification and identification method based on frequency domain characteristic. The research shows that the peak points of strips pattern’s frequency spectrum are on a line (feature line) which trough the center point of frequency spectrum, and the peak points of concentric circles pattern’s frequency spectrum are on a circle (feature circle) whose center is the center point of frequency spectrum. Firstly, this paper detects the peak points in the frequency spectrum of shoeprint images by fitting the probability density curve of the image’s frequency spectrum histogram. Then, referring to the idea of RANSAC algorithm, we define the curve of total energy of peak points and the curve of total number of peak points, then the possible feature line and feature circle can be determined by finding peaks of the total peak points energy curve and the maximum value of the total peak points number curve. Finally, the feature line detection is carried out by using corresponding frequency spectrum profile and the feature circle detection is carried out by using the connectivity of peak points near to the possible feature circle. The original images can be classified as the line type, the circle type, the line-circle type or the other type according to the detected feature line or feature circle. In addition, the interval and the direction of the lines on the original shoeprint images classified as the line type can be achieved according to the feature line parameters. These information can be used for the further classification of line type or retrieval and shoeprint matching. The experimental results indicate that the shoeprints classification method based on frequency domain features proposed in this thesis can efficiently realize the classification of the line type pattern and the circle type pattern.