Research On The Detection Of Petechia In Tongue Images

Tongue Diagnosis is one of the most important diagnostic means of Chinese Traditional Medicine that have been delivering for thousands of years; it plays an important part in the history of China. Recently, the development of computer science makes Chinese Tongue Diagnosis avoid the defect of inquantitativeness and subjectiveness,and makes automatic and objective diagnosis possible.Tongue petechia is an important pathological feature of Tongue Diagnosis. It plays a vital role in the judgment of several diseases. In this dissertation, image processing techniques are applied to detect the petechia in given tongue image.As the size of petechia is small, and its color is not determined, and nearly shows no information of shape and texture, which make it difficult in detection. For a long time, the research on it has not made much progress. In this dissertation, we have analyzed the characteristics of petechia, listed the possible interference, and proposed corresponding removal methods. For there are certain differences between petechiea, different detection methods are implemented on them. We have described the detection process in detail, and analyzed the potential problems in this dissertation, and the optimal method is chosen after experimental comparison.First of all, we preprocess the image. Both the saturation information of the reflective point and morphological erosion are utilized to remove the shadows around reflective point, which are also applied to remove the shadows on the tongue edge. In order to remove useless image pixels, smoothing and enhancing the image and then propose an improved method of extending the contrast. and the final experimental result is acceptable.During the process of big petechia detection, in order to remove small noise on the tongue, we compare several methods, and adopt the mathematical morphological open-close filtering method at last. Then we use the mark-based Watershed algorithm to segment the image. At last, we propose a method of removing selected region on the tongue edge based on the image reconstruction; and propose a method to distinguish the real petechia and the fake one. The detection results are admissible. During the process of small petechia detection, in order to remove the block shadow on the tongue, we adopt the Top-hat method. Then we propose a segmentation method based on gray histogram matching. About the identification of petechia, we first adopt the region-grow method, and then the hit-miss operation is implemented to remove the irregular regions and keep the petechia remain.The experiments prove that our methods are effective and practical.