Robust Face Recognition
|Project Title||Design & Development of an FPGA-based Multi-Scale Face Recognition System|
|Funding organization||ICTRnD Fund|
|Project Directors||Dr. Imran Naseem, Dr. Muhammad Mohinuddin, Dr. Husain Parvez|
|Expected Start date||September 2013|
|Budget||13.8 million PKR|
|Location||Signal Processing Research Group@PAF-KIET, Embedded Systems Research Group@PAF-KIET|
With increasing security threats, the problem of invulnerable authentication systems is becoming more acute. Traditional means of securing a facility essentially depend on strategies corresponding to “what you have” or “what you know”, for example smart cards, keys and passwords. These systems however can easily be fooled. Passwords for example, are difficult to remember and therefore people tend to use the same password for multiple facilities making it more susceptible to hacking. Similarly cards and keys can easily be stolen or forged. A more inalienable approach is therefore to go for strategies corresponding to “what you are” or “what you exhibit” i.e. biometrics. Among the other available biometrics, such as speech, iris, fingerprints, hand geometry and gait, face seems to be the most natural choice [A. Jain et al., 2004]. It is nonintrusive, requires a minimum of user cooperation and is cheap to implement.
Face recognition methods can be categorized as: (1) Holistic approaches and (2) Local features based methods. Holistic approaches such as Principal Component Analysis (PCA), Linear Discriminant Analysis (LDA) and Independent Component Analysis (ICA) treat the whole face as a single entity to derive face features. Local features on the other hand divide a given face into a number of spatially-local regions and then strive for the most robust local features. Local feature based approaches, such as Scale invariant feature transform (SIFT) and Linear Binary Pattern (LBP), have shown good performance compared to their holistic counterparts [A. F. Abate et al., 2007].
Wavelet transformation essentially decomposes an image into different frequency subbands. This transformation has shown some interesting results for the problem of image classification [A.Laine et al., 1993]. It is however believed that some frequencies are more discriminant than others and therefore a careful selection of these discriminant subbands is likely to improve the classification performance. Texture classification has been the main focus of these concepts. The task of identifying the most discriminant subbands for a given face recognition system is an open problem.
This project will be focused on the design and development of a high speed FPGA-based multiscale face recognition system using the Linear Binary Pattern (LBP) features. Special emphasis will be given to the algorithm design which can be efficiently mapped in Hardware. The LBP features primarily extract texture information of a face image. Wavelet decomposition of these features result in subbands encompassing low and high frequency components. These subbands carry useful information for classification, however some subbands are more significant than others and an intelligent selection of these discriminant subbands is likely to increase the overall performance of the face recognition system. The project is therefore aimed to identify the discriminant subbands for efficient and robust face recognition.
Increasing security threats in our society highlight the importance of the proposed project. Face is a non-contact biometrics and requires a minimal of users’ cooperation. Contact biometrics such as iris and fingerprints require a lot of cooperation on behalf of users and consequently fail when user is not willing to cooperate, for example one cannot expect a terrorist to present his iris to the security system. Therefore for security applications, by far, face is the best choice. Video surveillance is an efficient way of securing a facility. The proposed face recognition project will be available to be used in video surveillance. It will not only add value to the international research scenario but will also carry significant contribution to the society. The project will also be available as a stand-alone FPGA-based prototype solution that can be marketable through a startup company in domain of video surveillance.
Dr Imran Naseem has extensive experience with the state-of-art face recognition systems. His benchmark work was use of linear regression to solve the most challenging problem of occlusion in face recognition. He has also successfully proposed and designed robust regression based face recognition to encounter the difficult problem of illumination variation. Dr. Mohiuddin has extensively taught and worked on Xilinx FPGAs as target platform for VHDL (Very High Speed IC Hardware Description Language) based digital designs. Dr. Husain Parvez has extensively worked on design and exploration of FPGA architecture and related CAD tools. He has extensive industrial experience in the development and optimization of video related industrial applications.
Deliverables submitted to ICTRnD Fund
Details about Milestones and Deliverables
More details about achieved milestones and deliverables can be viewed in this document.