Face Detection under Occlusion or Rotation

• Main Authors: Chih-Hua Chang, 張之驊
• Other Authors: Shu-Yuan Chen
• Format: Others
• Language: en_US
• Published: 2011
• Online Access: http://ndltd.ncl.edu.tw/handle/70521896557501494433

碩士 === 元智大學 === 資訊工程學系 === 99 === Automatic recognition of human faces is a preliminary task in many applications such as security checks, human interaction systems, and face collection management. On the other hand, the prerequisite for a successful face recognition system is to detect faces in a given unknown picture. However, the task of automatic face detection in a complex background is difficult to cope with in particular for occluded or rotated faces, lighting distortions and non-uniform illumination. In this thesis, a face detector is proposed to resolve the problem. Our method consists of two phases: classification and verification. In the classification phase, a classifier using local block edge characteristics and support vector machine is first proposed. Sliding window approach using the proposed face classifier is then adopted to search multi-scale face candidates followed by mean-shift method to reduce duplicate face candidates so as to locate face candidates accurately. In the verification phase, 2-means algorithm is invoked locate face candidate region more accurately. The face classifier can then applied again to confirm whether a face exists. In this way the error of false alarm can be reduced. Various experiments prove the feasibility and effectiveness of the proposed method.