Study on Region-Based Forensic Image Retrieval

Forensic image database plays an important role in public security area. In order to narrow down the 'semantic gap' between the abundant semantic meanings of the query and the low-level image features, and to improve the precision of the content-based image retrieval (CBIR) for forensic images, this paper proposes a method which combines region semantic template with the ontology describing the relationship between region semantics in an image and the class of the image. In this method a number of objects are selected as representative components of the class. For every object, a set of sample regions are chosen to obtain the semantic template of the object, defined as the average feature of the sample regions. During the retrieval process, images containing same object as that in the query are selected, and then the ontology is used to further region the list to find the desired images. Experimental results prove the proposed method to be effective in forensic image database retrieval.