Research On Face Sketch-photo Hallucination Technology Based On Support Vector Regression

In practical application, face recognition is becoming more and more important,and face sketch-photo recognition has been a new branch of recognition technology.Due to significant differences between photos and sketches in texture, structure andinformation expression, face sketch-photo recognition depends heavily on facesketch-photo hallucination. Face sketch-photo hallucination in this paper refers to facesketch-photo synthesis and face sketch-photo enhancement, which can improve thequality of pseudo-sketch and pseudo-photo and increase the recognition rate in facesketch-photo recognition experiment. Therefore, research on sketch-photo hallucinationis of great importance.First of all, in this paper a new method of face sketch-photo synthesis based onsupport vector regression is presented. This algorithm combines supervised method withunsupervised method to synthesize pseudo-sketch and pseudo-photo, which takes intoaccount both the quality of synthesized image and the complexity of the algorithm. Theexperimental results demonstrate that the proposed method can achieve significantimprovement on both face sketch-photo recognition and perceptual quality.Furthermore, to solve the problems of blurring effect and blocking effect inconventional image, a new algorithm based on support vector regression is proposed toenhance the result of face sketch-photo synthesis. The final synthesized image isobtained by combining the initial estimate and the SVR based high frequencyinformation together to further enhance the quality of synthesized image. The validity ofthe proposed strategy was proved by face sketch-photo recognition and perceptualquality.Finally, an object assessment for massive face pseudo-sketch/pseudo-photo ispresented. The proposed evaluation algorithm includes two aspects: full-referenceimage quality assessment and no-reference image quality assessment. Full-referencequality assessment in this paper adopts the exiting visual information fidelity (VIF),however, the no-reference quality assessment needs to consider both the overall qualityof the image and the part details such as eyes, ears and so on. Therefore, theno-reference method combines the holistic features and local details. The experiment results indicate the effectiveness of the proposed no-reference method, and then validatethe significance of the proposed face sketch-photo synthesis and enhancementalgorithms.