The research is aimed at the development of an image processing system for classification of pathological area for medical images obtained from computed tomography (CT) scans. We proposed a novel semi-supervised image segmentation method based on the curvelet transform and SVM classfication. Firstly, through curvelet transform ultrasound images were decomposed into different directions and scales, the main distribution curvelet coefficients were extracted by cauchy model to reduce the algorithm time complexity, after inverse curvelet transform to obtaine a series of feature vectors from main distribution curvelet coefficients, then training samples and test samples were constructed; Secondly semi-supervised SVM classifier was designed, in order to reducing the weak classifier error rate, iteratively adjustment method was used to modify the SVM parameters, thus SVM strong classifier was constructed; Finally the expert manual tagging map were taken as reference standards, comparison with the existing method, experimental results shows that our algorithm is high anti-interference and has higher accuracy and effectiveness for ultrasound images pathological region segmentation.
Curvelet Denoising with Improved Thresholds for Application on Ultrasound Images
