Color Space Analysis Of Facial Image And Its Application In Disease Diagnosis

Facial diagnosis is an important part of traditional Chinese Medicine,mainly through the observation of human facial color changes to find the law of disease.With the development of science and technology,combined with computer to diagnosis is gradually becoming the field that many scholars try to study in depth,and try to objective and informative the Chinese Medicine Diagnosis.So far,for the facial diagnosis,there have been some research results.However,there are not unified evaluation standard and information standard,so still need to further explore.This topic is mainly for an input facial image,to find out the hidden meaningful characteristics,which have positive implications for disease analysis and can be used for health and disease analysis.Thus,the main contents of this paper include: facial image acquisition and pretreatment,facial image segmentation,facial color space analysis and feature selection,and disease analysis.In this project,the face image sample collection is mainly through the cooperation with the hospital,we put the self-developed equipment in the hospital for the collection.The obtained sample's label and the biochemical index are determined by the doctor.To the current position,the number of samples is collected more than 5,000 cases.In order to reduce the effect of the low sample size in the later period,this research mainly chooses the category in which the sample number is relatively large.The collected original images are color-corrected,and then the image segmentation is done,and the five skin blocks of interest are segmented for color analysis.In order to find out the color space distribution of different categories of samples,all the image samples in the faceted database were segmented to obtain the skin blocks,the skin blocks were projected into the color space,and the whole facial color distribution was fitted to the boundary.At the same time,the color distribution area of each category sample including disease and health was fitted,and the find out the distribution area of the base color.Finally,the density statistics of the Non-Base color distribution area is analyzed,and the representative color of the disease is found by clustering.The representative color as an index,extracted from the color feature vector for the next step of the study.Based on the representative color as an index to obtain the color feature vector,the related analysis of disease and health is done,which including dichotomous between health and various diseases,as well as multi-class experiments conducted on all samples.We choose different classifiers algorithms to carry out the experiments on two classification of the health and each kind of disease.The SVM classifier with better classification effect is selected to classify at the end,and the average correct rate is more than 80%.From the experiment,it is proved that the color analysis is meaningful which color characteristic extraction based on representative color.In order to more practical applications,we try to use logical regression and SVM decision tree for multi-classification analysis.The recognition rate of multi-classification can achieve relatively high recognition rate in individual diseases.In practical multi-classification applications,it is also proved that facial color features are meaningful for disease diagnosis.