Face Recognizability － best face shot candidate for surveillance system
| Autor: | Kai-Fang Cheng, 鄭凱方 |
|---|---|
| Rok vydání: | 2005 |
| Druh dokumentu: | 學位論文 ; thesis |
| Popis: | 93 It is an urgent desire for researchers to uplift the accuracy in human face detection and recognition. The main bottleneck lies mainly on the quality of input images which definitely drastically affect the accuracy of the system. In this thesis, the information of facial feature is adopted to determine the direction of face rotation and then find the best shot of faces in a surveillance video stream so that the quality of input images can be improved. The features to be extracted include skin color information and edge information. Skin color information can be obtained by analyzing the skin color distribution in YCbCr color space and edge information can be detected by applying Sobel edge detector. Moreover, three strategies are proposed to determine the three directions of face rotation based on the feature image. The first strategy uses the collection of vertical feature projection. The direction and the angle of face turning can be determined by analyzing the distribution of vertical projection histogram. The second strategy uses a novel model called radial template to detect the presence of face rotation. This template is designed to find the angle of center-rotated objects. According to the characteristics of skin detection and edge extraction, the extracted feature will be stable under this kind of template. The third strategy is to determine the presence of bending or lifting of faces based on the relationship of feature area and neck position. By integrating the three strategies and the geometry of eye position and judgment of symmetry, the best shot of frontal face image can be identified which can be employed in later face recognition task. Experiments were conducted on various video images containing faces. Experimental results verify the feasibility and validity of our proposed approach in determining the most frontal faces in video sequences. |
| Databáze: | Networked Digital Library of Theses & Dissertations |
| Externí odkaz: |
|