Research On Tongue Detection And Tongue Segmentation In Open Environment

Tongue diagnosis is an important diagnostic method in the Traditional Chinese Medicine(TCM)which has caused widespread concern worldwide due to its noninvasive and convenience.Traditional tongue diagnosis depends a comprehensive analysis of the physiological and pathological characteristics of tongue body,which is subjected to the clinicians' knowledge and experience,and influenced by environmental light and other factors.Fortunately,digital tongue diagnosis system based on the computer image processing technology is developed to solve this problem,which achieves the objectification,quantification and automation of tongue diagnosis,also improves the efficiency and utilization of tongue diagnosis.The digital tongue diagnosis system is the only way to develop modern tongue diagnosis.Currently,digital tongue diagnosis system mostly collect tongue image in a closed environment with a fixed lighting condition,in order to obtain high quality tongue images.However,with the development and popularization of mobile devices such as smart phones,the image acquisition using mobile devices in open environment has gradually become a new research direction.But the attendant problem is that the image acquitted in open environment is vulnerable to light intensity,complex background and other factors.Therefore,before the analysis,the detection and segmentation of tongue body from the image is needed.In this paper,a new method of tongue detection and segmentation is proposed for the open environment.For the image acquisition in open environment,the main work of this paper contains the following three parts:1.An improved image preprocessing method is proposed.Firstly,the median filter is performed on the image,in order to remove the noise and smooth the image.Then,a novel color correction method is propose,on the basis of the traditional gray color correction algorithm,the convergence value of the algorithm is adjusted to meet the characteristics of the tongue image that the average value of each channels has a different proportion,and the linear correction of the traditional gray world algorithm is changed to the nonlinear correction,specifically,gamma transformation is adopted,which avoids the intensity of resultant image too large or too small.Finally,the brightness of image is corrected.The experimental results show that the method is effective and has practical value.2.On the basis of the preprocessing image,a new method of tongue detection is proposed.The idea of the algorithm is to split the image and then detect,which can improve the detection speed and meanwhile get the basic contour of the tongue body.For the former,the image is separately binarized by using OTSU,hue threshold and RGB component difference method,after binarization,the morphology operation is utilized to smooth the connection domain,and then perform“and”operation on the three results of binary image to get the final segmentation result.For the tongue detection approaches,we achieve the purpose of tongue detection by extracting the features of each connected domain and training random forest classifier.3.On the basis of tongue detection,we can get the basic contour and position of the tongue body.Therefore,an improved method of tongue image segmentation based on active contour model(ACM)is proposed.The improved gradient vector flow active contour model(GVF Snake)is utilized for the further segmentation,which is compared to the traditional Snake model,traditional GVF Snake model and the distance regularized level set evolution(DRLSE).The experimental results show that the tongue detection and segmentation method proposed in this paper effectively solves the problem of tongue detection and segmentation in open environment that has better robustness and accuracy.