PROJECT TITLE :

Dual-transfer Face Sketch-Photo Synthesis

ABSTRACT:

Law enforcement and criminal investigations, among others, rely heavily on the ability to identify a sketched face from a face image dataset. There is no need for artists to hand draw sketches in an intelligent system that relies on photo-based face sketch generation. Conventional face sketch-photo synthesis procedures, on the other hand, tend to produce sketches that are consistent with the artist's drawing style. Because identity-specific information is frequently overlooked, identity verification and recognition fail to meet expectations. Conventional methods fail to recover identity-specific information, and this research explains why. Then, we present a new dual-transfer face sketch-photo synthesis framework that includes both an inter-domain transfer mechanism and an intra-domain transfer procedure. Transferring a regressor from the test image into the sketch image ensures that common face structures are recovered during synthesis thanks to the inter-domain transfer. Learning and transfer of a mapping between images and sketches across identities prevents identity-specific information from being lost during synthesis in intra-domain transfer. By employing an ad hoc information splitting method, the two procedures can easily be combined. The suggested framework is implemented using both linear and nonlinear formulations. A study using the face sketch database from the Chinese University of Hong Kong shows that the suggested framework produces more identifiable facial structures and achieves better face identification in both the photo and sketch domains compared to the present state of the art.