| Currently people pay more attentions to clinical infrared thermal image diagnoses technology, since it can be performed both functional and structural test. But the simple thermal image processing methods in current software prevent its widely usage from medical area.In this thesis, we start from the basic knowledge about infrared radiation and medical infrared thermal image. We design some thermal image processing methods according to their character, such as denoising and background extraction in preprocessing step, the identification of region of interest, statistical analysis for symmetry parts of body as well as long-distance diagnoses and help system. These enable the doctor to assess thermal image in an automatic approach.In order to improve the accuracy in medical thermal image diagnoses, some new methods are proposed. (1) The noise in thermal image mainly follows Poisson distribution, so we use PRESS (Predictive Sum of Squares) optimal threshold in wavelet domain to remove the noise. (2) We use the improved fast algorithm to separate the background from the image, and morphological steps were performed to fill the possible holes created by an aggressive thresholding. (3) We delimitate the regions of interest for body parts and use statistical measure and isothermal analysis to compare the symmetric ROIs. We also propose a new statistical approach using distances between histograms of comparable ROIs. (4) Canny edge detection, Hough transforms and other technologies are used to extract the boundaries of the breasts. Unsupervised learning technique is applied to classify each segmented breast. According the distribution of nose temperature, we use spatial searching method to extract the contour of nose.We design the file management and help system to overcome the deficiencies in current software, and the long-distance diagnoses system in software will make the consultation convenient for experts in different places, so we can make good use of expert resource. |
PDF Full Text Request