Dissertation > Excellent graduate degree dissertation topics show

Research on Multiresolution Nonrigid Registration Based on Improved Mutual Information

Author: SunYaLan
Tutor: GaoXiePing
School: Xiangtan University
Course: Computer Software and Theory
Keywords: Breast X image Computer-aided diagnosis Comparative analysis Improved mutual information model Wavelet-based mutual information model Multiscale elastic registration
CLC: TP391.41
Type: Master's thesis
Year: 2008
Downloads: 51
Quote: 0
Read: Download Dissertation

Abstract

Mammary census carried out based on the images of the breast X is the early detection of breast cancer lesions, an important means of reducing mortality. Images of the breast X-based computer-aided diagnostic technology is the current hot research abroad is difficult. The current mainstream of computer-aided diagnosis doctors in clinical review films commonly used contrast back, this means the introduction of computer-aided diagnosis is formed based on the pieces of breast X-ray contrast analysis based on a single X-ray film; method. Key step for the elastic matching two shape, neither the same size of the breast X image of the comparative analysis of the use of multiple X-rays to resolve this problem, scholars proposed method can be divided based on the control points on the the method of registration and registration method based on the similarity measure. Both face challenges, however, the former because of the particularity of the breast X image, in addition to the nipple position and no obvious control points used to calculate the elastic transformation parameters; latter when the transformation parameters more easily into local extremum. Automatically achieve accurate registration of the images of the breast X is a very challenging task. The images of the breast X-based computer-aided diagnosis system module division, summarizes the research and the contribution made by foreign scholars in this field, the emphasis on the comparative analysis method. On the basis of previous studies, automatic elastic registration method combined with improved mutual information model and wavelet multi-scale analysis. The method does not require pre-select control points, according to the local registration transformation parameters extracted global registration control points, in order to achieve the overall elastic registration. Simple mutual information registration method as a similarity measure considering only a single pixel grayscale information, and does not consider the spatial information. This paper summarizes the previous several improved mutual information models, including normalized mutual information model, the model of mutual information and gradient information, the area mutual information model and symbiotic mutual information model, proposed a new wavelet-based mutual information model and compare the performance of the three wavelet in this model. Contrast with the previous four methods improved experiment, the results prove that the same as the model of mutual information and gradient information, the proposed method is superior to the method the Josien [33] proposed, and has good robustness. This method is applied to this multi-scale elastic registration procedure in the local alignment.

Related Dissertations

CLC: > Industrial Technology > Automation technology,computer technology > Computing technology,computer technology > Computer applications > Information processing (information processing) > Pattern Recognition and devices > Image recognition device
© 2012 www.DissertationTopic.Net Mobile