| Summary: | 碩士 === 國立臺灣科技大學 === 材料科學與工程系 === 102 === The larynx is an important breathing channel and phonatory organ of human. The symptoms of cancer tumor are very important for the doctors to diagnose the patients for laryngeal disease. At present, the patients' larynx is examined by laryngoscope, upon personal diagnosis by the doctors. In order to remedy the doctors' probable omissions and careless mistakes in the identification process, an automatic tracking system for laryngeal cancer tumor area in the laryngoscope image sequence is designed in this research. The laryngeal cancer tumor area is framed and tracked continuously and displayed in the video, so as to assist and remind the doctors of the laryngeal cancer tumor in the video.
Based on several groups of endoscope image sequence samples, the image samples are processed and screened according to color space features and textural features. First, the textural features are extracted from the entire image by using gray level co-occurrence matrix, combined with the statistical distribution diagram to determine the four effective features, including contrast, homogeneity, dissimilarity and inverse difference moment normalized, from the six groups of textural features. Hence, normal image can be separated from low resolution and fuzzy images completely. The low resolution and fuzzy images are then filtered using the optimal features of contrast as indicator.
The image is cut to 20×20 sub-images by using partial texture analysis, and the judgment conditions are established to delete the useless black region information in the image, so as to reduce the amount of calculation of co-occurrence matrices. Six texture parameter differences between general tumor and cancer tumor are determined and tabulated for reference. The improved homomorphic filtering is used for global and partial processing to eliminate the uneven illumination of image and to strengthen the contrast. The position of laryngeal cancer tumor in the first image is found by color feature and image morphology and shielding filtration calculation. Finally, the self-developed adaptation threshold tracking method is used to track the tumor position in subsequent image sequences continuously. The cancer tumor symptom area is circled continuously by contour capture in the image sequence, so as to assist the doctors with clinical diagnosis and evaluation.
|
|---|
