基于人脸画像的伪照片合成及应用分析-pseudo - photo synthesis and application analysis based on face portrait.docx

ABSTRACTFace sketch plays an important role in identifying the criminal suspect in fields of law enforcement and counter-terrorism. However, the difference between face sketch and real face photo is enormous and it is impossible to directly identify face sketch by traditional face recognition technology. In addition, human visual sensibility of sketch and photo is different and it is limited to recognize face sketch by artificial recognition. It will afford a great help for law enforcement and the counter-terrorism that utilizing the method based on image transformation to synthesize face pseudophoto based on input face sketch.Aiming at the specific problem of transforming face sketch into face pseudophoto, this paper proposes a face sketch based pseudophoto synthesis and amendment method. Thisalgorithmcontainsthreesections:thefacepseudophotosynthesizing, enhancement and local amendment. Firstly, the initial estimate of face pseudophoto is synthesized by local eigentransformation method. Secondly, a simple and effective initial estimate enhancement method is proposed based on features of the input sketch. Finally, a control-point based local deformation method is proposed to amend the local synthesis error of face pseudophoto. The framework of this paper is summarized as follows:① This paper first introduces the academic research significance and realisticvalue of face pseudophoto problem research, analyzes face sketch recognition of domestic and abroad research status, and emphasizes on analysis and summary of pseudophoto and pseudosketch synthesis based on image transformation. It would lay theoretical foundation for the subsequent research by analyzing the difficulty point of research subject.② This paper analyzes the key technology and theory of face pseudophotosynthesis method based on image transformation, introduces the classic global eigentransformation method based on global transformation and the local geometry preserving method based on local tran