Research On The Digitalization Technologies Of "Differentiation Of Syndromes By Observing Eyes"

The digitalization of Chinese medicine diagnoses has becoming a hot topic in TCM andcomputer communities with recent progress of electronic and computer technology."Differentiation of syndromes by observing eyes" is one kind of diagnosis by observing, whichobserving the change of the white of eye and the state of its vessel. The digitalization of"Differentiation of syndromes by observing eyes" can help to build an objective andquantitative diagnostic standard for Chinese traditional medicine. In this dissertation, the keydigitalization technologies of "Differentiation of syndromes by observing eyes" are studied.The major contributions are included as follows:1) A method is proposed for color correction in eye image, which adjusts von Kriescoefficients according to the type of image region. It is showed that the von Kriescoefficients of Munsell ColorChart vary with the color sample under different illuminants.The image is preprocessed by gray-world algorithm at first. Then image regions arelabeled with four types: illuminant-color, illuminant complementary color, near neutraland others. For the region of first two types, von Kries coefficients are adjusted and vonKrise adaptation are applied. The near neutral region is corrected by white balance. At lastimage region of all types are combined together. This method can reduce the color cast ofeye images.2) A method based on adaptive edge tracking is proposed for the automatic vessel extractionin eye images. Sclera-conjunctiva region is isolated by optimal thresholding segmentationat first. The scan lines in sclera conjunctiva region are searched for the initial trackingpoints. Second-order derivative of Gauss filter is used to determine the edge points, andthe adaptive low-pass directional filter is employed to detect the orientation of edge. Thetracking vector is adjusted in both direction and length dynamically during trackingprocedure. This method can restrain the background noise, track vessels which havediscontinuities in curvature and diameter. The branches are also processed properly.3) A fast marching algorithm based on adaptive mesh refinement is proposed to solve shapefrom shading problems in the eye image. The mesh is dynamically refined according to thesurface curvature, and the refine level is under control. Hash table is used to quickly locateneighboring nodes. The causality condition is proven theoretically for first orderbackward/forward difference. This algorithm can reduce the error in high curvature region,improve numerical accuracy and intensify noise immunity.4) A fast marching algorithm based on isoline-tracking is proposed. All nodes in narrowbands are picked up one time to avoid the sorting. Then, the values of nodes in expandingband are updated by upwind scheme once or twice in different orders. The expansion ofisolines is also recorded and tracked. The time complexity is proven to be O(N). Thisalgorithm can be extended to multisource problem and the time complexity is still O(N).This algorithm can improve numerical efficiency and accuracy, solve SFS and heightrecovery problems in eye image, and is much faster than fast sweeping method incomplicated situations.5) An experimental system of "Differentiation of Syndromes by Observing Eyes" is builtbased on above technologies. The system consists of two parts: construction of eye imagedatabase and automatic diagnosis. It's the first time to explore and verify the digitalizationtechnology of "Differentiation of Syndromes by Observing Eyes".