Dissertation > Excellent graduate degree dissertation topics show

Research on Key Technologies of Iris Image Preprocess and Feature Extraction

Author: WangJunHui
Tutor: ZhangDaPeng
School: Harbin Institute of Technology
Course: Computer Science and Technology
Keywords: Iris diagnosis Gamma correction Statistics eigenvectors SOM network Relative entropy
CLC: TP391.41
Type: Master's thesis
Year: 2010
Downloads: 43
Quote: 0
Read: Download Dissertation

Abstract

Chinese and Western medicine found through long-term study of the iris of the human eye, the human iris tissue structure with the occurrence of the diseases of the body, the development of changes, and according to the characteristics of different changes in the iris structure with different body organ lesions corresponding. This study from the iris image texture changes in the region, given the different diseases and normal texture corresponding feature definition and template to identify different diseases. This paper curling round, the sun radiation the ditch and normal texture regions as the object of study. The full text reads as follows: First, a detailed introduction to the iris image preprocessing. This paper proposed an iris image adaptive Gamma correction method based on the the feature maps regulation positive. Combined with the characteristics of the iris image, a distance Gamma correction method, and the use of the simulated annealing algorithm adaptively determine the Gamma value. The chord detection pupil coarse positioning, fine positioning round deformable template iris image. The use of non-concentric with polar coordinate transformation, the annular iris normalized to a fixed size rectangular region. Second, given the clear and obvious pattern of the three types of area curling round, sun reflection ditch and normal class definition. The paper also proposed a training sample extraction method based on sliding window. This method will be preprocessed iris image is divided, according to the above definition different iris region, we need to elect training samples. In this paper, using the gray level co-occurrence matrix (GLCM), fractal and statistical methods such as Hu invariant moments to describe the texture characteristics of the sample. Texture features for training samples directly extracted texture features based on the first wavelet transform and feature extraction methods. Third, the proposed method based on self-organizing neural networks iris feature selected. SOM (Self-Organizing Feature Map) network clustering results showed three types of samples gathered to a different output on the neurons, having a good degree of distinction between them, we defined for these three samples and is reasonable. We use the relative entropy analysis and evaluation of the selected feature, and given a relatively reasonable interpretation of the results of feature selection. Finally, application support vector machine (SVM) and SOM two classifications, while the the unknown marked nervous system diseases, digestive system diseases and healthy people sample classification test. The results show that the proposed method to extract the disease characteristics curling round, the sun radiation ditch and better distinguish normal region, and laid the foundation for the preliminary correct diagnosis.

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