Research On Measurement And Recognition For Man Face Skin Texture

With the improvement of the living standard, people more and more concern cosmetology and nursing care of man face skin. Therefore analysis measurement and recognition of skin texture become a significant research task. The traditional measurement based on observing of eye does not conform to the demand of people. The pattern recognition system for measuring man face skin texture based on statistical properties analyses of image texture on the basis of now available theory and the literature is researched.Firstly the significance of evaluating impersonally and quantificationally of face skin texture in the field of cosmetology and nursing care through the human body skin physiology mechanism is analyzed, the defect of traditional measurement is put forward and the pattern recognition and the texture analysis technique included their research developing actuality is introduced.Secondly, several texture statistical properties analysis techniques, included spatial gray level co-occurrence matrix, center matrix, run length measure and max-min value measure, are thoroughly studied and actual texture meaning of every feature in this some kinds of statistics means is qualitatively analysed. Furthermore, actual physical sense of texture feature of spatial gray level co- occurrence matrix is proved through a group of testing images. Then, based on the traditional BP network algorithm, a new improved TFBP network algorithm (Transfer Function Error Back Propagation) as classification and recognition algorithm of the artificial neural network pattern recognition is put forward.In the end, the system for collecting man face skin image based on co-focused microscope and CCD camera is designed and preprocessing of skin image and calculating of the characteristic value with VC program language is programmed. Adopting synthetically spatial gray level co-occurrence matrix, center matrix, run length measure and max-min value measure to extract features of skin texture is proved more valid for classifying and recognizing skin texture. During the training experiment of skin texture image, 15 series of specimens of five types of skin image are selected as the data and the training has been underway to TFBP network withthe neural network toolbox of MATLAB. Finally six types of man face skin texture images are identified, and the result that classification recognition was correctly lead to 83.3% demonstrates that measurement and classification recognition system of man face skin texture is feasible and its theory analysis is correct.