Dissertation > Excellent graduate degree dissertation topics show

The uncertainty of the data mining algorithm design

Author: LiXiaoLi
Tutor: WenJun
School: University of Electronic Science and Technology
Course: Computer System Architecture
Keywords: Uncertainty Data Mining Fuzzy Clustering Image Segmentation
CLC: TP311.13
Type: Master's thesis
Year: 2011
Downloads: 49
Quote: 0
Read: Download Dissertation

Abstract

Traditional data mining has solved the problem of huge amounts of data with poor knowledge. However, it is just suitable for accurate data. Because uncertainty is an inherent feature of objective things, the result of data mining may not be right without taking into account of uncertainty. With uncertain data increasing, we need Data Mining that takes into account of uncertainty urgently. Uncertain Data Mining can mine in uncertain data. In the research field of Uncertain Data Mining, the Cluster Analysis and application of the fuzzy uncertain data is the most widely used technology. Fuzzy uncertainty means that there is no clear extension of things. This thesis focuses on the FCM algorithm and analyzes its role in the field of image segmentation. The essence of Image segmentation is to cluster similar image pixels together and separate different image pixel, which is as same as the essence of Cluster Analysis. Because of the ambiguity of image which is caused by error imaging and the character of human vision, and the need for an automated image segmentation algorithm, fuzzy clustering algorithms as an unsupervised algorithm can meet these demands. Although the application of FCM algorithm in image segmentation has been very extensive, FCM itself has many shortcomings, such as the large amount of calculation, the low computation speed, the sensitivity to initial value, the poor character of convergence, the trend to fall into local minimum, the large amount of iteration and so on. Besides, FCM algorithm always gives a result for any giving data and initial value, while it can not determine the quality of clustering results.This thesis presents an improved FCM for image segmentation. To solve the problem of low speed, this thesis quantifies the image data firstly. For gray image, the thesis uses the approach of eigenvector which takes the gray-level statistics as the weight; for color image, this thesis uses the approach of quantifying color sets which takes the statistics of quantized color set as the weight. Then, the thesis uses weighted method to calculate the compressed data. This approach not only can guarantee the accuracy of segmentation result, but also can improve the speed. Then this thesis uses weighted subtractive cluster analysis to cluster the quantized image data approximately, which can not only determine the maximum number of clusters automatically, but also can get initial cluster centers corresponding to each cluster. These initial centers are the data points that have the largest density, and they are closed to the true cluster centers, so they can avoid the situation of unsuitability of the initial value, large amount of iterations, and the problem of local minimum. Finally, the thesis uses Cluster Validity based on Possible Distribution to identify the validity of clustering results and to obtain the best cluster result.In this thesis, experimental results show that the proposed algorithm not only guarantees the accuracy and validity of image segmentation, but also has faster speed and smaller amount of iterations than the FCM algorithm.

Related Dissertations

CLC: > Industrial Technology > Automation technology,computer technology > Computing technology,computer technology > Computer software > Program design,software engineering > Programming > Database theory and systems
© 2012 www.DissertationTopic.Net Mobile