Research On Cluster Based Shoeprint Retrieval Algorithm

Shoeprint is one of the important evidences of criminal investigations. For the amount of shoeprint images acquired from crime scences is very huge, it is one of the most urgent tasks that the criminal technology faces to locate effectively the same type shoeprints to the one left in the crime scene in the dataset with large amount of shoeprint images. The existing shoeprint image retrieval algorithms don’t consider the effect of both the correlation between images and the same pattern shoeprint images on the retrieval result, which makes the retrieval result lack of semantic information to some extent. A shoeprint image retrieval algorithm based on clustering is proposed to improve the performance of the shoeprint image retrieval algorithm in this thesis.The main works of this thesis are as follows:1) A shoeprint image retrieval algorithm framework based on clustering is proposedAn algorithm frame of shoeprint image retrieval algorithm based on clustering is proposed according to the defects of the current algorithms. The whole retrieval process is divided into two phases which include the automatic clustering stage and reterival stage. The similarity between two images in the dataset and the effect caused by the same pattern shoeprint images from the dataset are added to the process of the traditional retrieval algorithms to improve the performance. Experimental results on three kinds of test dataset show that the proposed algorithm is very effective.2) A K steps stabilization based automatic clustering algorithm of shoeprint is proposedThe automatic clustering method is proposed to part the shoeprint sets effectively based on the analysis of the distribution of the shoeprint images in the feature space. The core idea of the proposed algorithm is to find the margins between classes and use them to part the shoeprint set. Result Experimental results on two kinds of public available databases and one real shoeprint database composed of 5792 images have shown that the proposed algorithm outperforms the state of the art clustering algorithms on common clustering evaluation measures. The precision and F-measure of the proposed algorithm on the real shoeprint database are about 99.68 and 95.99 percent respectively.3) A shoeprint image retrieval algorithm based on data set clustering is proposedThe proposed algorithm considers the local relationship between two dataset images and the global relationship among the same pattern images, and the make the ranking scores satisfy the following three conditions:(i) the consistency of the similarity between the ranking score and feature similarity, (ii) the consistency of both the difference of ranking scores and the difference of their feature similarity values,and (iii) the ranking scores of the shoeprint image from the same cluster in the dataset is closed to each other. The experiment results show that the value of MAP and NDCG have reached 79.84% and 87.15% respectively, and the performance of proposed algorithm is improved about 20% and 12% respectively compared with traditional algorithm.4) A shoeprint image retrieval algorithm based on query images clustering is proposedMultiple shoeprints with the same pattern left in the crime scene is considered in this work. Multiple shoeprints with the same pattern should have the same ranking scores, therefore, a ranking score object function is construted according to the similarity relationship between multiple crime scenes images and the images in the dataset. The experimental results show that the value of MAP and NDCG have reached 84.39% and 91.89% respectively, and they are improved about 30% and 20% respectively compared with the image retrieval algorithm with only one scene image.The semantic information that the ranking scores of similar images or images from the same cluster shoud be close is incorporated into the process of shoeprint image retrieval. The experimental results on public available test databases and real shoeprint databases show that the proposed algorithm has a better performance and is well correlatated with human opinions. The proposed algorithm has been applied to the practical shoeprint restrieval systems.