3 April 2008 A special purpose knowledge-based face localization method
Author Affiliations +
This paper is concerned with face localization for visual speech recognition (VSR) system. Face detection and localization have got a great deal of attention in the last few years, because it is an essential pre-processing step in many techniques that handle or deal with faces, (e.g. age, face, gender, race and visual speech recognition). We shall present an efficient method for localization human's faces in video images captured on mobile constrained devices, under a wide variation in lighting conditions. We use a multiphase method that may include all or some of the following steps starting with image pre-processing, followed by a special purpose edge detection, then an image refinement step. The output image will be passed through a discrete wavelet decomposition procedure, and the computed LL sub-band at a certain level will be transformed into a binary image that will be scanned by using a special template to select a number of possible candidate locations. Finally, we fuse the scores from the wavelet step with scores determined by color information for the candidate location and employ a form of fuzzy logic to distinguish face from non-face locations. We shall present results of large number of experiments to demonstrate that the proposed face localization method is efficient and achieve high level of accuracy that outperforms existing general-purpose face detection methods.
© (2008) COPYRIGHT Society of Photo-Optical Instrumentation Engineers (SPIE). Downloading of the abstract is permitted for personal use only.
Ahmad Hassanat, Ahmad Hassanat, Sabah Jassim, Sabah Jassim, } "A special purpose knowledge-based face localization method", Proc. SPIE 6982, Mobile Multimedia/Image Processing, Security, and Applications 2008, 69820M (3 April 2008); doi: 10.1117/12.777546; https://doi.org/10.1117/12.777546

Back to Top