| This dissertation proposes a novel image segmentation algorithm of the Markov random field theory in the fuzzy domain and a novel edge detection algorithm based on fuzzy logic and inference rules, and discusses the application of the fuzzy Markov random field image segmentation algorithm to breast ultrasound image segmentation. This research proposes a novel automatic classification system for masses on breast ultrasound images using statistical methods.; We propose a fuzzy Markov random field segmentation algorithm based on the newly defined local energy, which takes into consideration the influence of the neighborhood system, and implements it for breast ultrasound image analysis. Experiments demonstrate that the proposed algorithm is better than the stochastic EM/MPM image segmentation algorithm, the scale space image segmentation algorithm, and the segmentation approach using a function-based parameter to weight the components in the MRF model in the literature, based on their test images. Experiments also show that the proposed algorithm is better than the K-means clustering, the fuzzy C-means clustering, and the standard MRF image segmentation algorithms. It works well on noisy breast ultrasound images. The experiments (statistical feature extraction/selection, mass lesion classification) based on the segmentation images demonstrate that the proposed approach results in a high classification accuracy of masses on breast ultrasound images, and indirectly proves the accuracy of the segmentation results. The proposed approach will be valuable for assisting radiologists in the diagnosis of breast cancer.; While being a staple of image segmentation algorithms, edge detectors are sensitive to noise. We propose a novel fuzzy edge detector based on fuzzy if-then inference rules and edge continuity. It employs the fuzzy maximum entropy principle to determine the fuzzy membership functions automatically. The proposed edge detection algorithm is compared with the Sobel and Canny edge detectors. The proposed fuzzy edge detector does not need parameter setting as the Canny edge detector does; it preserves image details, is robust to noise, and works well under high-level noise situations, unlike other edge detectors. Related issues in fuzzy edge detector design are discussed. The experimental results demonstrate the superiority of the proposed method to existing ones. |
