Research On Personalized Facial Image Synthesis Based On Epistemic Logic

Personalized facial image synthesis has a very wide range of applications in the fields of criminal investigation, film and television production, entertainment and education, and it has attracted more and more attention of many researchers. However, previous studies limited to specialists and researchers in the fields of computers. Sadly there is little cross-disciplinary collaboration. It is quite difficult to solve personalized facial image synthesis problem from a purely technical point of computers because human face is non-rigid in shape and it is similar to other individuals, and the detect methods are too susceptible to disturbances. Current research on personalized facial image synthesis is still in its infancy, and there were very few mature applications.The emergence of epistemic science not only provides a platform for the integration of computers, philosophy, psychology and other disciplines, but also provides some new direction for solving the personalized facial image synthesis problem. The problem of how to integrate cognition into the design of computer systems and how to improve computer program with philosophical speculative thought will be a topic worthy of further study in the future for a long period of time. Starting from the application of personalized facial image synthesis, we tried to design the computer system with the guide of epistemic logic, and run through the ideas of epistemic logic in our programming, so as to escape the current predicament of the research of personalized facial image synthesis.After considering every link of personalized facial image synthesis system from the perspective of epistemic science, a personalized facial image synthesis method based on epistemic logic was presented in this paper. First of all, face images normalized through illumination compensation based on symmetric block and grayscale adjustment algorithm based on age. Secondly, the feature points of face images were extracted by the Active Shape Model algorithm. Thirdly, face contours were morphed by image deformation algorithm based on feature line pairs, and face textures were transplanted by wavelet image decomposition and reconstruction techniques. Thus customization of a personalized facial image was realized which contained different ages, different weight status and different standards of living. Moreover the idea of "human intelligence and machine intelligence were combined and man played a leading role" run through our design and human-machine cooperative work environment was bulit through human-computer interaction.Experimental results showed that this method can effectively synthesize personalized facial images, and the simulation results under different conditions had greater discrimination which also were rather realistic.The main work and innovations of this paper are as follows:1. After analysing current illumination normalization methods from the perspective of epistemic science, an illumination compensation algorithm based on symmetric block and grayscale adjustment algorithm based on age were put forward, which improved the grayscale normalization method of face images. Most of the current illumination compensation algorithms such as histogram equalization or spherical harmonic wavelet basis function mainly treated the images as a whole, and they might not fit rules of human being’s cognizance. In order to improve the effects of pre-processing, the symmetrical areas in the darker side were compensated by the background light on the lighter side, and good practical results were achieved. In addition, the skin color characteristics of different ages were considered and the change rules were summed up which were used to guild the design of grayscale adjustment algorithm. The adjusted face images had certain age characteristics which provided a good base data for subsequent programs.2. Human facial feature extraction methods widely used were summed up, and the theory and running process of ASM were specifically focused on which was improved for personalized facial image synthesis in this paper. At present there were four main methods to extract face features:method based on geometry features, method based on statistical features, method based on frequency domain features and method based on mixed features, while ASM was one of the most widely used classic algorithms with good precision and robustness. However, there were some error which would influence integral perception of the face images, especially on the localization of some key parts such as the mouth, eyes, etc, just because the ASM algorithm was sensitive to the initial state and it’s ending condition was global contraction. To solve this problem, feature points trimming mechanism were introduced in which the error of ASM could be partly corrected through the specialists ’confirming. At the same time,22feature points which can characterize hair and hairline were added to the original68feature points of images in FG-NET database, so as to better meet the requirements of realistic application.3. Current mainstream image morphing techniques were summarized and their advantages and disadvantages were compared. After considering this technology from the perspective of epistemic science, the image deformation algorithm based on feature line pairs was improved and face contour morphing algorithm based on epistemic logic was brought forward. The core of the morphing algorithm was settings of the line pairs which decided the effect of morphing. According to the characteristics of face images, non-uniform characteristic line pairs selection method was proposed in which the selection of line pairs tilted to key parts such as the mouth, eyes, etc, thus the effect of image morphing was promoted. Three key parameters of the algorithm were analyzed from the perspective of cognition, and their cognitive meaning was made clear which was validated by experiment.4. The basics of wavelet transform were introduced, and the principle of wavelet decomposition and reconstruction of two-dimensional image was analyzed in detail. After that a face composite method based on epistemic logic was put forward. To solve the shortage of direct replacement method and enhance the effect of transplanting, an improved method was presented in which the low frequency part of sample image was filtered by Butterworth high-pass filter and added to that of the test image. Considering the phenomenon that hair and hairline played an important role in age judging, a hair replacement algorithm was designed in which the hair and forehead of the sample image was replaced to the corresponding area of test image. Through the replacement algorithm, the composite image became more realistic.5. A personalized facial image synthesis system based on man-computer interaction was designed guided by the idea of "human intelligence and machine intelligence were combined and man played a leading role". In the system, experts’wisdom and computer’s high speed and capacity were organically combined through the cooperative work of experts and computers. Computers were liberated from imaginal thinking which they were not good at, and just acted as a super secretary and assisted experts to work. The problem that the ASM algorithm was inaccurate and there was some error in age estimation algorithm was corrected by experts, and the ending of system was decided by the experts’assessment. Those aspects of human-computer interaction made the system more efficient and effective.Finally, the paper was summarized, and the directions for further research in the future is proposed after analyzing three shortcomings of the design.