Research On Shoeprint Images Multi-label Clustering Algorithms

The shoeprint image is one of the most important evidences of the crime scenes.Automatic classification of shoeprint images plays an important role in improving the efficiency of cases detection.The existing shoeprint image cluster algorithms focused on single-label cluster.However,the distinctions among different classes of shoeprint images are small,that is to say,there is no obvious isolation with different classes.At the same time,a shoeprint image often contains one or more different types of patterns,so different shoeprint images contain the same type pattern.Therefore,a shoeprint image can be classified according to its containing patterns.Based on analyzing the characteristics of shoeprint image dataset,we propose the multi-label clustering algorithm for shoeprint image.And the main works are as follows:1)A multi-label clustering algorithm for shoeprint image based on improved Fuzzy c-means(FCM)is proposedBy improving the membership matrix of FCM,the algorithm considers the relationship between image and clustering center,at the same time,considers the relationship between images.The algorithm starts from the high density data and takes advantage of the relationship among image and classes to achieve shoeprint image multi-label clustering.The relationship is determined by the improved membership matrix.The F-measure experimented on the actual shoeprint image dataset has reached 79.09%.2)A multi-label clustering algorithm for shoeprint image based on random walk is proposedThe algorithm firstly proposes to apply the random walk algorithm to optimize the shoeprint image dataset similarity matrix.By means of clustering from each image,merging categories and remarking the multi-label image,the algorithm realizes multi-label clustering of the shoeprint images.The algorithm not only considers the relation between the images,also considers the relation between classes.The F-measure experimented on the actual shoeprint image dataset has reached 78.34%.3)A multi-label clustering algorithm for shoeprint image based on probabilistic latent semantic analysis(PLSA)is proposedThe probability distribution matrix among shoeprint images and latent semantic topics,which learned from shoeprint image element semantic vocabulary by PLSA model,is used to establish the relation between the images.After single-label clustering and multi-label clustering two processes,the algorithm finishes multi-label clustering of the shoeprint images.The F-measure experimented on the actual shoeprint image dataset has reached 73.61%.