【作者】 马媛媛；

【导师】 孙君顶；

【作者基本信息】 河南理工大学 ， 计算机应用技术， 2010， 硕士

【摘要】 纹理分析是计算机视觉和数字图像处理中的一个重要的研究课题,而如何获得纹理特征是其中的重要环节。本文主要围绕图像特征提取、BP神经网络技术和遗传算法在图像分类与检索中的应用展开,首先介绍了图像检索技术的发展概况、关键技术和研究现状,在系统讨论纹理特征提取的过程中,采用了基于纹理灰度共生矩阵作为图像特征。其次,为了提高图像检索效率,本文将基于遗传算法改进的BP神经网络引入到图像分类中,结合纹理特征进行图像分类识别,并用于图像检索中。本文的主要工作和创新如下:1.综述了基于纹理分类的图像检索技术,介绍了纹理的定义、分类、分析方法和纹理分析的应用。鉴于纹理的分析是基于纹理的图像检索技术的重点,论文分别分析总结了统计分析、结构分析、模型分析和频谱分析四类纹理分析方法。2.提出了一种新的图像分类算法,首先利用灰度共生矩阵方法提取出图像的纹理特征,然后结合遗传算法优化的BP神经网络进行网络训练和样本分类。本算法避免了无关样本图像进行图像特征匹配的过程,有效地节省了图像检索的时间开销。实验结果表明,将图像的纹理特征和改进的BP神经网络相结合,有效准确地实现了对给定图像的分类,缩小了查找图像的范围,提高了图像的查准率,并且很大程度上减少了图像检索的匹配时间。更多还原

【Abstract】 Texture analysis is an important research topic of computer vision and image processing. How to get the texture feature is the most important step. The exploratory research begins with the application of the texture feature extraction, BP neural network and genetic algorithm in image retrieval. Firstly, the development, key technology and research status of image retrieval technology are introduced. Gray level co-occurrence matrix is adopted as the image characteristics while discussing texture feature extraction. Secondly, in order to improve the efficiency of image retrieval, BP neural network is presented for image retrieval, which is also optimized by genetic algorithm.The main research work and innovation of this thesis are given as follows.1. Retrieval technology based on texture feature is summarized, the definition, classification, analytical method and application of texture are introduced and analyzed. The statistical analysis, structural analysis, model analysis and spectrum analysis for texture are also discussed in detail in the paper are researched and summarized.2. A new image classification algorithm is proposed. Firstly, the GLCM is adopted to describe texture feature of an image, combine the neural network optimized by the genetic algorithm for network training and sample classification. It avoids the image feature matching process of the independent image, and ruduces the time complexity for image retrieval. Experiment results show that it can classify the image effectively and accurately combining texture characteristics and the improved BP neural network. It shows that the proposed algorithm is effective and feasible in image classification and image retrieval.更多还原