一种新的基于关键子块的图像检索算法

A Novel Image Retrieval Method Based on Keyblock

为了将有效的文本检索技术应用到图像检索中,结合人眼视觉特性及方块编码的思想,提出了一种基于图像关键子块的检索算法.即首先利用图像方块编码的思想将图像预先分成互不重叠的子图像块,然后利用方块编码的思想,根据块的灰度差对这些子图像进行独立地编码,这些子图像的方块编码构成的块不仅能有效的描述图像的纹理内容,而且可以反映图像的形状分布和边缘分布.以此来定义图像的关键子块.最后借助文本检索技术来实现图像检索.同时,考虑到不同类型关键子块在图像中出现的频度对检索效果的影响,又提出了相应的改进算法.实验结果表明,该算法是有效的.

Content-based image retrieval（CBIR） came from text information retrieval（TIR）, but many technologies in TIR can＇t be used in CBIR because of the feature difference of both. In order to effectively use existing test information retrieval methods in content-based image retrieval, a novel image retrieval method based on keyblock is presented incorporating the different sensitivity along variant directions of human visual system and block truncation coding （BTC）. Firstly, the image is divided into equally sized blocks from which the image keybolcks are extracted by use of the method of BTC according to the different direction of the texture distribution. Then, the method of TIR is used in the image retrieval. After that, an improved method based on the weighted histogram is proposed because the frequency of different kinds of keyblocks in the image has different influence on the image. The method can achieved a higher efficiency because of integrating spatial distribution information and edge distribution information into image descriptor. Experimental results have shown that the proposed method has sound and robust retrieval performance especially for the images with the abundant texture information and edge information.