Abstract
Image segmentation can effectively segment the interested objects from the image, and it often was used in pattern reorganization and medical image processing etc. Otsu’s thresholding (OTM) method is one of widely used image segmentation methods. However, OTM cannot successfully give a proper threshold when the standard deviations or the numbers of data in different classes are quite different from each other. In this thesis, “adaptable threshold detector” (ATD) is proposed to overcome the drawbacks of OTM. The ATD minimizes “within-class standard deviation” (WCSD) as the criterion, which considers the standard deviation of group, the quantity of data, and the group interval within a group to be factors in deciding the optimal thresholds. The cervical cancer is one of women’s common diseases and its incidence and mortality rates are ranked prior top in the woman’s common diseases. In order to effectively improve the accuracy and reduce the cost of image segmentation, this thesis proposes a nucleus and cytoplasm contour detector for cervical smear image, which is called “nucleus and cytoplasm contour” (NCC) detector. First, the NCC detector adopts ATD to detect the contour of cytoplasm on a cervical smear image. After that, the NCC detector employs the maximal gray-level-gradient-difference method to extract the contour of nucleus from the extracted cytoplasm region. Transmitting data to people or medical image among health care professionals, providers, and organizations has become popular via the Internet. However, the illegal attackers or hackers can easily grab, duplicate, or revise the data on the Internet. The reversible image hiding method can embed the secret data into cover image for protecting the secret data that be stolen by the illegal attacker. However, reversible image hiding method can recover the original cover image without any distortion while extracting the secret data. In the thesis, a high payload frequency-based reversible image hiding method is proposed. In the proposed method, the secret data are embedded into the coefficients in a high-frequency band in frequency domain. In a medical image, “region of interest” (ROI) is a region which contains important information and must be stored without any distortion. This thesis hence proposes an ROI-based image hiding method. In this method, the ATD is applied to automatically segment ROI. Next, the secret data embeds in non-ROI by an irreversible image hiding method and in ROIs by a reversible image hiding method to increase embedding capacity.