【作者】 张熠转；

【导师】 马培军；

【作者基本信息】 哈尔滨工业大学 ， 计算机科学与技术， 2007， 硕士

【摘要】 随着网络技术、多媒体技术、数据库技术的发展和互联网的不断普及,图像信息应用越来越广泛,人们对图形、图像等多媒体数据的需求也越来越强烈。基于语义的图像检索不仅方便于用户的使用,还准确地体现出用户的意图,因此是图像检索发展的必然。而图像的标注字能够较好的表达图像的语义内容,它能够缩小图像的高级语义和低级视觉内容之间的差距,因此图像语义的自动标注方法也正逐渐引起人们的重视。本文通过分析图像语义自动标注的相关技术,深入探讨和研究了图像视觉特征的提取方法。在颜色特征提取方面,利用HSV(Hue, Saturation, Value)颜色空间进行非等间隔量化,并构造一维特征矢量,用累加直方图表示图像的颜色特征;在纹理特征提取方面,针对不同纹理特点分别采用了基于共生矩阵的统计纹理分析和基于小波变换的频谱纹理分析两种方法予以实现。通过对以上特征的提取,实现了融合多特征的图像检索。针对不同特点的图像,融合不同的图像特征,并采用不同的相似性度量方法,提高了图像检索的准确率。在检索结果的基础上,采用基于实例的方法实现了对图像语义的自动标注。这种方法的基本思想是,把带有标注的训练样本集当作一种标注经验,在提取出示例图像的视觉特征后,从经验库中检索出与之视觉相似的图像,并且通过模仿这些例子图像的标注,对图像实施标注。通过大量的仿真验证,表明该语义自动标注方法不仅能够将图像的视觉特征转化为图像的标注字信息,可以有效的应用于基于语义的图像检索,而且克服了手工标注费时费力的缺点,为用户的使用带来了极大的方便。更多还原

【Abstract】 With the development of web-technique, multi-media, database-technique and unceasing popularity of the Net, the using of image is more and more popular, and the requirements for multi-media data such as graphs and images are more and more intense. The semantics-based image retrieval can not only be convenient for users, but also deliver their intentions exactly, so it is the inevitable way of image retrieval. The annotations, which are able to reduce the gap between high-level semantics and low-level visual content, can well express the semantic content of images, so the automatic annotation of image semantics is being paid more and more attention.This paper has a further exploration and study of visual feature extraction depending on analyzing correlative technology of the automatic annotation. According to the HSV(Hue, Saturation, Intensity) color space, the work of color feature extraction is finished, the process is as follows: quantifying the color space in non-equal intervals, constructing one dimension feature vector and representing the color feature by cumulative histogram. Similarly, the work of texture feature extraction is obtained by using co-occurrence matrix or frequency analysis based on wavelet transform depending on different characteristics of images. Depending on the former algorithms, image retrieval based on multi-feature fusion is achieved.Fusing different image features and using different similarity measures depending on different characteristics improves the accuracy of image retrieval. At last, on the basis of retrieval results, an example-based method is introduced to annotate images automatically. The training data are stored as the annotation experiences. In order to annotate a new input image, visual similar images are retrieved from the database. Annotation words can be derived from imitating the annotation examples of the retrieved images.A large number of simulations show that the semantic annotation of image semantics is not only able to change visual features of images into annotations, which are very useful for semantics-based image retrieval, but also overcomes the shortcomings of manual annotation, which is time-consuming and strenuous, and provides users with great convenience.更多还原