The integration of wavelet analysis and support vector machine for face detection | |
| Author | Vo Duc My |
| Call Number | AIT Thesis no.ISE-09-22 |
| Subject(s) | Image processing Interactive video |
| Note | A thesis submitted in partial fulfillment of the requirements for the degree of Master of Engineering in Mechatronics, School of Engineering and Technology |
| Publisher | Asian Institute of Technology |
| Series Statement | Thesis ; no. ISE-09-22 |
| Abstract | My thesis is focused on the task of detecting faces in arbitrary images. Face detection is the first and most important step of face recognition. If we can’t locate exactly faces in any scene, the recognition process will be very complicate or impossible. Furthermore, face detection has many interesting applications: It can be part of surveillance system, security, or a video-based computer/machine interface. Face detection was based on a classification algorithm to separate face patterns from non-face patterns. There are many difficulties for face detection problem, for examples: the dimensionality of sub-images is usually high, number of non-face sub-images is extremely large and their distribution is very irregular, the probability distribution of face patterns is very difficult to model. Therefore we present a novel approach to solve these difficulties. The method comes from the integration of Two-Dimensional Wavelets Transform and Support Vector Machine (SVM). This integration brings many advantages to detect exactly human face. Support Vector Machine is extremely good at solving above difficulties. The main ideal of Support Vector Machine technique is to separate the classes with a surface that maximizes the margins between them. It can be classified the very large number of samples (several millions samples) in very high dimensional space. It can discriminate between samples that belong one of two classes with a good performance. In order to have the highest performance, we need to implement feature extraction step. Facial features extraction makes SVM model classifying more accurate and easier. In this step we apply the Two Dimensional Wavelets Transform to decrease the vectors redundancy and guarantee a set of the most Specific and useful features for training SVM model. All of the sub-images must to be preprocessed before training SVM model. Preprocessing process includes two periods: illumination adjustment and histogram equalization. Illumination adjustment reduces the lightning and shadow. Histogram equalization compensates differences of brightness. The experimental results shown in the thesis demonstrated the high accuracy of this detection system. |
| Year | 2009 |
| Corresponding Series Added Entry | Asian Institute of Technology. Thesis ; no. ISE-09-22 |
| Type | Thesis |
| School | School of Engineering and Technology (SET) |
| Department | Department of Industrial Systems Engineering (DISE) |
| Academic Program/FoS | Industrial Systems Engineering (ISE) |
| Chairperson(s) | Afzulpukar, Nitin V.; |
| Examination Committee(s) | Manukid Parnichkun;Dailey, Matthew N.; |
| Scholarship Donor(s) | MOET scholarship; |
| Degree | Thesis (M.Eng.) - Asian Institute of Technology, 2009 |