Research On Purposive Vision And Its Application In Skin Symptom Recognition

Recently purposive vision has been a study of interest in the field of machine vision. The system of purposive vision should possess an active sensibility and be based on certain tasks or purposes. In view of the importance of purposive vision and the practicability of intelligent skin diagnoses, the author makes an in-depth research in the field of machine vision by specially studying the skin-symptom recognition and researching for analyzing the skin symptom image with purposive machine vision. The author studies the manipulation of purposive vision, constructs the corresponding algorithm modules and their chip implementation methods.The research contents of this project includes building the elementary theoretic modules for the purposive machine vision (skin image processing), designing IP-core"MV-01"—a high-performance microprocessor with a MCS-51 kernel and taping out chips with the function of the purposive machine vision and multi-processors embedded on. The main work and achievements are listed as below:1. Preprocessing of skin micro-imagesThe author develops a dyadic wavelet method to calculate the amplitude of the image after the second-rank wavelet decomposition and to eliminate the local maxima by a dynamic threshold, which effectively clears up the disturbance from other things on the skin such as the fine hair without affecting the useful information. The reconstructed image is the denoised image which preserves the image edges and details. This method could be used in the denoising of skin symptom image, and other fields such as the gene array. A method of Skin Micro-image Adaptive Segmentation transforms the skin image from the RGB to HIS model and uses the S (saturation) to express the image. The method simplifies the work from three dimensions to one dimension. Furthermore, it retains the color information. Methods of the segmentation of different skin micro-images are automatically selected, which is based onσ,standard difference of S value. Whenσis larger, max-variance method is used to determine thresholds. Otherwise, corrected-mean method is used to determine thresholds. This method could be used in the segmentation of a large-scale standard difference. It improves the objectivity and accuracy of the segmentation.2. Feature extraction,feature selection and symptom recognition of skin images The symptom of skin micro-images is extracted from the skin image background, and then features will be sorted out according to certain segmentation. According to the...