Crime Scene Investigation Image Retrieval using a Hierarchical Approach and Rank Fusion

摘要

A two-phase hierarchical approach for Crime Scene Investigation(CSI) image retrieval was proposed. In the first phase, the approach uses multi-class classifier to classify the class label of the query image. In the second phase, according to the classified label, a smaller dataset or sub-database only including images with the same label as the query image is picked out from the image database. Then Query By Example(QBE) approach is used to retrieve images from the sub-database. Furthermore, fusion results are returned by different representations for the same query based on rank fusion model. Pilot experimental study was carried out on a database with 18,004 CSI images, which are collected from the actual cases and divided into 10 semantic categories labeled by public security officer. Experimental results showed that retrieval accuracy based on the proposed approach is encouraged.