【作者】 王静；

【导师】 付先平；

【作者基本信息】 大连海事大学 ， 计算机应用技术， 2008， 硕士

【摘要】 随着数码设备、网络及多媒体技术的发展,人们在工作、学习和日常生活中不断地产生大量的图像数据。因为数字图像呈现了爆炸式的增长方式,如何合理而又高效地组织海量的图像数据、结合图像低层特征,将数字图像进行分类和检索是目前的一个研究热点。本文所做的研究工作包括以下几个方面:1、改进的基于颜色的图像分类方法研究。特征提取是基于内容检索的关键技术,它决定了图像分类性能的好坏。而颜色特征是图像物理特征中最直接的视觉特征,相对于其它特征,颜色特征非常稳定,对于图像的平移、尺度、旋转等变化不敏感,具有很强的鲁棒性,而且颜色特征计算简单。因此,在本论文所实现的图像分类系统中,采用HSV颜色模型对颜色空间进行量化后,分别对图像进行了全局代表色和分块代表色的提取,并通过实验对比了两种特征提取方式的图像分类准确率。2、基于支持向量机的图像分类方法研究。支持向量机能够较好的解决小样本学习问题,其目标是得到现有信息下的最优解而不仅仅是样本数趋于无穷大时的最优值。本文在详细论述了支持向量机算法的基础上,采用一对多SVM方法,为每一类图像分别构建一个分类器,应用序列最小化(SMO)分类算法,实现了基于支持向量机的图像分类系统。3、图像分类中相关反馈方法的实现。由于图像的低层视觉特征与高层概念的相关性较弱,且在图像分类系统中用户的主观性比较强,因此本文将相关反馈技术加入到图像分类系统中,通过用户与系统的实际交互过程进行图像分类。由于相关反馈过程可以看作模式识别中的二分类问题,因此本文将SVM算法用于相关反馈的学习和分类过程中。实验结果显示在图像分类系统中加入相关反馈技术能有效的提高图像分类的准确率。更多还原

【Abstract】 With the development of digital equipment,network and multimedia technology, more and more digital images are created in people’s working,learning and daily life. Because digital images appearing explosive growth mode,it is a hot research topic that how to organize the mass image data effectively and how to classification and retrieval the digital image with the image low level characteristids.The content of this paper are listed as follows:First,research on image classification method based on color.Feature extraction is the key technology of content based image retrieval,it decide the classification performance.Since color feature is relatively robust to background complication and independent of image size and orientation,it is widely used in image feature representations.This paper uses HSV color model to quantify color space in the image classification system,and abstracts image feature by the representative color of traditional global histogram and image block separately.The accuracy of these image feature extraction methods for classification is also discussed.Second,research on image classification based on support vector machine.Support vector machine can solve small sample learning problem very well.The aim of learning is not only to get optimal values when samples tend to be infinite,but also to get the optimal solutions under current conditions of information.The detailed algorithmic of support vector machine,using One-Against-Rest SVM method to construct classifiers for each image class,and using sequential minimal optimization(SMO) classification algorithm to implements image classification system based on support vector machine are discussed in this paper.Finally,the implementation of relevance feedback method in image classification is introduced.Because of the weak correlation of low-level features and high-level conception of images and different subjective perception of the system users,this paper introduce relevance feedback technology into image classification system,classify images in the process of the interaction between users and system.The process of relevance feedback can be regarded as a branch of classification in pattern recognition, SVM algorithm is used in the process of relevance feedback learn and classification. The experiment results show that introduce relevance feedback technology to image classification system can improve the precision effectively.更多还原