【作者】 李巧玲；

【导师】 张卫国；

【作者基本信息】 西安科技大学 ， 计算机应用技术， 2011， 硕士

【摘要】 随着计算机网络和多媒体技术的迅速发展,图像资源越来越丰富,传统的文本关键词检索方法已经不能满足图像信息的检索要求,基于内容的图像检索技术(Content-Based Image Retrieval,CBIR)应运而生,并成为图像领域研究的热点问题。本文重点探讨基于颜色特征和形状特征的图像检索方法。首先,本文论述了目前国内外基于内容的图像检索系统的发展和研究现状,研究了基于内容的图像检索的关键技术,包括图像颜色、纹理、形状以及空间特征的提取,相似性度量,相关反馈机制和检索性能评价技术。其次,对于形状特征的提取,提出了一种基于改进的不变矩特征的图像检索算法(ISE算法),在传统的七个Hu不变矩特征的基础上加入了离心率特征,把这八个特征作为图像的形状特征并进行归一化之后进行图像检索。实验数据表明,基于ISE算法的图像检索具有对图像的平移、旋转、缩放、扭曲的不变性,较好地保持了形状的一致性,尤其对于目标明确的图像具有良好的检索效果。再次,对于颜色特征的提取,针对全局颜色直方图丢失了空间信息的缺点,对图像进行分块,并选用HSV颜色空间且将颜色量化为72维,统计每个分块的颜色直方图并选取像素点数目最多的那种颜色作为分块的主颜色。将示例图像与检索图像对应分块主颜色距离的累加和作为两幅图像的相似性距离进行图像的检索。最后,研究了融合颜色和形状两种特征的图像检索方法,设计了一个基于分块主颜色和改进的形状不变矩特征的图像检索系统,并能够实现两种特征的单独检索功能和综合检索功能。实验结果表明,如果图像中的目标物体比较明显,或者想要检索到包含物体的平移、旋转、缩放以及形变的图像,使用形状不变矩特征的检索效果更好。用户可根据图像的特点给两种特征分配权值进行综合检索,比传统单一特征检索的效果要好。更多还原

【Abstract】 With the development of network and multimedia technology, image resources become more and more abundant, the traditional method which based on text keyword is unable to fulfill the demand of image retrieval. Content-Based Image Retrieval which becomes one of the hottest research topics in the field of image has emerged at the right moment. This paper mainly investigated the methods of image retrieval which based on color and shape.Firstly, a general overview of the development and research actualities of the content-based image retrieval system are introduced, and its key technology is researched, including the extraction of the image features (such as color, texture, shape and spatial relationship), the similarity measurement, relative feedback and retrieval performance evaluation.Secondly, as for the extraction of the shape feature, an image retrieval algorithm (ISE) is proposed based on modified moment invariant features, in the algorithm, eccentricity is joined in the seven traditional Hu moment invariants. Experimental data show that image retrieval based on ISE algorithm has some advantages of translation, rotation, scaling and distortion invariance of image, and has good affect in maintaining the consistency of the shape, and especially has good retrieval results to the target specific images.Thirdly, as for the color feature extraction, because of the loss of space information of global color histogram, image is divided into small pieces, HSV color space is chosen and the color is quantized to 72-dimensions. Color histogram of each partition is counted, and the color which contains the largest number of pixels is dominant color of the partition. The implementation of image retrieval is through counting similarity distance which is the cumulative distance of the corresponding blocks between the sample image and database images.Finally, in order to design and implement the content-based image retrieval prototype system by synthesizing the feature of dominant color of the partition and modified shape invariant moment, the image retrieval method integrated with color and shape features are researched. The system can retrieval images based on single feature or color and shape features. The experimental results show that image retrieval based on invariant moment has better retrieval efficiency while the image has specific targets or the user want to retrieval images which include translation, rotation, scaling or distortion targets. According to the characteristics of the image, users can retrieval images by giving different ratio to the two features. The retrieval precision of fusion multi-feature is much higher than traditional single feature retrieval.更多还原