基于机器学习的医学超声图像边缘检测算法研究-计算系统结构专业论文.docxVIP

西南科技大学硕士研究生学位论文第 西南科技大学硕士研究生学位论文 第 PAGE II 页 西南科技大学硕士研究生学位论文第 西南科技大学硕士研究生学位论文 第 PAGE III 页 Abstract Medical ultrasound imaging technology for its excellent performance has become an important aid tool in clinical diagnosis. Edge detection is a key step in medical ultrasound image processing, and its test results will directly affect the doctors diagnosis of the disease. Speckle noise of the ultrasound image make the traditional edge detection algorithm processing effect not very satisfactory. In addition, Sobel operator and other traditional edge detection algorithms have the limitations of parameter settings, which restricted their use in the ultrasound image. Image edge detection problem can be seen as the classification problem of edge points with non-edge points. So it can be studied using machine learning classification thoughts. The machine learning algorithm in the field of medical image edge detection is at the initial stage. In order to study the above-mentioned problems of noise and parameter settings, this paper makes the medical ultrasound image as the research object and uses AdaBoost algorithms combination with the decision trees to study the edge detection problem. Creating a training set is the first step in machine learning algorithm, and it needs to annotation image. But some lacks such as spend long time, a strong sense, and workload exist in manual image annotation. This paper proposes an automatic annotation ultrasound bladder image integer DCT-based algorithm, which makes the machine learning algorithms more practical and promotional. The main work of this paper is as follows: （1） Doing medical ultrasound images experiment with the classical edge detection operators including gradient operator, Canny operator, Sobel operator, Laplace - Gauss operator, analyzing and comparing their advantages and disadvantages. （2） Proposing an automatic image annotation ultrasound bladder intege r DCT-based algorithm. Through a series of steps such