| Summary: | 碩士 === 國立中正大學 === 資訊工程研究所 === 87 === In this thesis, we propose a new breast tumor diagnosis system. The input images in our system are the ROI images. The ROI images are then applied by our segmentation algorithm and segmented to the tumor regions and surrounding tissues. Cooperating with the segmentation algorithm, three powerful features are extracted from the ROI images, which are variance contrast, auto-correlation contrast, and wavelet coefficient distribution distortion. To classify the images, we construct an MLP and train the MLP using error-back propagation algorithm with momentum. With the three features we proposed as inputs of the MLP, the breast tumor images are then classified quite well.
|
|---|
