الفهرس 
IntroductionOnly 14 pages are availabe for public view
 |  | from | 89 |  |  |
| | | from | 89 | | |
Abstract
Abstract
One of the challenges in face recognition is to match two faces with different modalities. Face sketch is widely used in crime investigation to help in identifying the criminal. In the face sketch recognition systems, a collection of face photographs (obtained from the suspect’s database) and sketches (drawn by a forensic artist depending on the description of some eye witnesses) are matched together. This thesis provides several contributions in face sketch recognition. The first contribution provides an unsupervised method for face photo-sketch recognition by synthesizing a pseudo-sketch from a single photo. It is divided into two main steps namely: edge detection and hair detection. The unsupervised method significantly improves visual quality of the synthesized sketch with respect to state-of-the-art methods. The improvement in synthesizing takes a very small execution time compared to the state-of-the-art methods. Moreover, it improves the recognition accuracy when matching viewed sketches (sketch is drawn by looking at a photo of the suspect or the suspect himself) with photos. It also improves the matching forensic sketches (sketch is drawn by a forensic artist depending on the description of an eye witness) with photos.
The second contribution provides a mean feature method for face photo-sketch recognition by synthesizing a pseudo-sketch from a photo or synthesizing a
iii
pseudo-photo from a sketch using training sets of photos and sketches. It improves the synthesized sketch and the recognition accuracy vs. some state-of-the-art methods. The third contribution provides an improvement for the synthesized photo by fusing the proposed mean feature method with a nonlinear method. The fourth contribution provides an improvement for the recognition rate using viewed sketches with 96%recognition rate at rank 1 using a local feature detection method (Scale Invariant Feature Transformation(SIFT)). The fifth contribution provides an improvement for the recognition rate using forensic sketches; the recognition rate is 57% at rank 50