Facial Component-based Representation For Matching Facial Composite Sketches To Photos

Facial composite sketches are useful tools of determining the identity of criminals for law enforcement agencies. Law enforcement agencies usually post facial composite sketches in public places and in media in hopes that a citizen who recognize the subject will come forward to provide a clue. This method of identifying suspects is slow and tedious and does not leverage the mugshot database in the police office. Automated facial composite sketch to photo matching can leverage the police mugshot database and help the law enforcement agencies apprehending suspects in a timely fashion.The key problems in matching facial composite sketches to photos are describing the invariant feature of faicla composite sketches accurately and matching facial composite sketches to photos in a fast way. Face composite sketches and photo of the same subject may differ in appearance due to the change in image modality, so the existed method cannot describe the invariant feature of facial composite sketches accurately. Because the exhaustive matching strategies are adopted by existed methods to match composite sketches to photos, it is very time consuming when match facial composite sketches against large mugshot database. In order to make automatic facial composite sketches to photo matching be applied to reality, this paper does research on problems which influence the performance of matching facial composite sketches to photos. The main work of this paper is as follows:① This paper first introduces the academic research significance and realistic value of matching facial composite sketches to photos, analyzes domesitic and abroad research status of matching facial composite sketches to photos, and the difficulty point of research subject.② Targeting the problem of describing the invariant feature of facial composite sketches, this paper analyzing four invariant feature descriptors, namely, Multiscal Local Binary Pattern(MLBP) descriptor, Histogram of Orientation Gradient(HOG) descriptor, Scale Invariant Feature Transform(SIFT) descriptor and Gradient Location and Orientaion Histogram(GLOH) descriptor. Evaluate the performance of these four descriptors in representing the invariant feature of facial composite sketches. The experimental results show that GLOH is good at describing the invariant feature of facial composite sketches and is less compute complexity.③ For addressing the problem of matching facial composite sketches to large mugshot database, a facial component code table-based method for narrowing down a mugshot database is proposed. The proposed method consists of two phases, namely, establishment of the FCC table database and narrowing down the mugshot database. In the first phase, a FCC table database is established, where Gradient Location and Orientation Histogram(GLOH) is used to describe the invariant feature of facial composite sketches. In the second phase, six FCCs(FCC of hair, eyebrows, eyes, nose, mouth, and jaw) of a composite sketch are retrieved in the FCC table database and a narrow mugshot database can be obtained. The experiments demonstrate that the proposed method can not only narrow down the mugshot database efficiently but also preserve the target photo with high probability.④ This paper introduces the GLOH into the framework of component-based representation for matching composite sketches to photos, and does research on a facial component gradient location and orientation histogram feauture based method for matching facial composite sketches to photos. The experimental results show that the method can match facial composite sketches to photos efficiently.