Study Of Identification And Classification Methods Of Shoe-Prints Based On Blocks

Footprints and shoe prints are very important and most easily collected evidences in criminal case investigation, and become more useful in serial and parallel. Since every local police department has different footprint management methods and only adopts manual operation, footprints can not be used as effectively as fingerprints in the investigation of criminal cases. It is necessary to find a kind of image processing method under this background, and to make use of computers to carry out shoe prints identification and classification automatically.In view of above background, shoe print classification must be carried out on the actual shoe print pictures to analyze picture correctly. Basing on the research works in image processing field that have been done in recent years in Dalian Maritime University, for the purpose of the automatic classification, the algorithms of classification methods based on blocks are brought forward as following: The original image are normalized first. Then the normalized picture is divided into 13×5 blocks and each block has 64 × 64 pixels. The blocks which include the outline of the edge and the background part are picked out by comparing the normalized picture with the standard outline of the shoeprint model. Basing on the first step, the LOG edge detection is carried out and the threshold is designed, blocks that include little or no texture pixels are deleted. The initial classification is made according to the Laws energy characteristic quantity calculated from the convolution of the template and the shoe print image. After the first step classification, a co-occurrence matrix of gray-scale and a co-occurrence matrix of the close dimensions are constructed based on Wavelet Transform. Taking the Haralick statistics as the characteristic quantity of the two co-occurrence matrixes, a new characteristic matrix is re-constructed, and the cluster analysis is carried out. The final classification is accomplished by the algorithms based on the classification results achieved from the above steps.Randomly sampled experiments with 500 shoes prints photos were carried out ant the results show that the classification algorithm is effective and the classification correctness ratio is high enough.