Research On Illuminance Compensation And Image Registration In Tongue Image Analysis

In the past two decades,the objectification of traditional Chinese medicine has caused increasing research attention.Research on the utility of visible illuminance images of the human body surface for medical diagnosis purposed is also emerging.As the coral approach of the visual perception of traditional Chinese medicine,the ”tongue diagnosis”has been extensively studied in the fields of acquisition,processing and analysis of tongue images.The thesis studies the high-performance preprocessing in tongue image processing and analysis.The existing tongue image analysis mainly focuses on imaging device,feature extraction,and tongue image classification.The preprocessing problem in tongue image analysis has been less seriously researched.Specifically,the illuminance interference and image registration of the tongue image have not received extensive attention.In the current research,the problems of tongue image illuminance interference and image registration are often ignored or dealt with in a simple way,lacking detailed research and effective solutions.The thesis focuses on the study of tongue image illuminance compensation and tongue image registration and proposes effective solutions.First,the thesis discusses the problem of tongue image illuminance interference and proposes an illuminance compensation method.The current research on tongue image illuminance interference mainly focuses on the removal of tongue image reflection points,and there are relatively few studies on tongue image illuminance drift.The issue of illumination drift of tongue image will have a huge impact on tongue image analysis.The thesis points out the linear drift problem of tongue image texture characteristics with the image illuminance,and proposes a compensation method for this drift problem.The proposed compensation method is based on the image Hodge decomposition model,which decomposes the image into the sum of the two components of illuminance and texture.Based on the image Hodge decomposition model,the thesis establishes a linear regression model between the tongue image Gabor texture feature and the illuminance,which explains the linear drift problem between the tongue image Gabor texture feature and the illuminance.Numerical experiments on the real tongue image database verify the correctness of the linear regression model.The experimental results show that the linear drift problem is solved by the illuminance compensation method proposed in the thesis.Moreover,based on the tongue images after compensation,the accuracy of the tongue image classification is improved compared with the tongue images before compensation.Secondly,the thesis studies the problem of tongue image reflection points and propose a corresponding removal method.Different from the illuminance drift,the tongue image reflection point is a high-frequency signal.The thesis proposes a class of nonlinear filters,which can simultaneously solve the problems of illuminance drift and reflection points of the tongue image.Specifically,the thesis proposes to use the wide line detector(WLD)filter response image replacing the original image as the underlying image for Gabor texture feature extraction.The preprocessed Gabor texture feature based on the WLD filter is robust to the interference of image illuminance drift,and at the same time eliminates the local highlights caused by the moist reflective points on the tongue surface.Therefore,in the term of describing tongue image coarseness,the preprocessed Gabor texture feature is more accurate than the existing Gabor texture feature.In the preprocessed Gabor texture features,the tongue image coarseness becomes more significant as a pathological information indicator,and the accuracy of tongue image classification based on the preprocessed Gabor texture features is also improved.Thirdly,the thesis discusses the tongue image registration problem and proposes an effective realization approach.The existing tongue image registration is mainly to establish the mapping on several positions between the tongue images,and is vulnerable to the interference from deformation of the tongue image.The research establishing dense and robust registration mapping between tongue images has not been studied yet.The thesis proposes a method of tongue image registration based on conformal mapping.In the proposed method,the tongue images are registered to the same standard shape.Based on the registered tongue image,we extract the average tongue and the eigen-tongue of the tongue image.The average tongue of each class shows some typical visual characteristics of the class,confirming some empirical observations and intuitive understandings in the practice of TCM diagnosis.The eigen-tongue indicates the functional partition within the tongue image area.The feature extraction based on the eigen-tongue has the advantages of high dimensionality and informativeness,thereby improving the performance of tongue image diagnosis.Finally,the thesis studies the conformal mapping algorithm for two-dimensional tongue image registration.The current research on conformal mapping mainly focuses on the registration between three-dimensional surfaces.This category of methods faces many difficulties when applied to the two-dimensional planar tongue image.Specifically,the existing three-dimensional conformal mapping method cannot overcome the orientation problem when applied to the tongue image.The thesis proposes a tongue image orientation method based on boundary alignment,in which the tongue image orientation problem is transformed into the alignment problem of a key point on the boundary.In addition,the thesis proposes a fast conformal mapping algorithm to make conformal mapping suitable for large-scale tongue image registration problems.Compared with the three-dimensional conformal mapping method,the proposed conformal mapping method can achieve tongue image registration more accurately,and the feature extraction based on the conformal mapping proposed in the thesis has a higher accuracy in tongue image classification..In summary,the thesis discusses two types of image preprocessing problems in tongue image analysis.The two proposed preprocessing methods can be employed as effective tools for tongue image processing and analysis.