基于多邻域空间分布的图像检索

Image Retrieval Based on Multi-Neighborhood Spatial Distribution

由于灰度共生矩阵及其改进算法存在计算复杂、且对于纹理分布信息缺乏计算以及忽略了图像相关特性等缺点,导致对于图像纹理的有效信息缺乏很好的描述,为此提出了一种新的纹理特征用于图像检索。该算法首先结合图像中像素的统计信息,针对不同的邻域范围提取图像的邻域相关矩阵,然后在此基础上构造多邻域空间分布特征用于图像检索。分析表明,该算法所提取的纹理特征计算量小,复杂度低,并且由于将纹理的结构特征和统计特征有效地结合起来,所以对图像的空间纹理分布特征可以较好地描述。为了证明新算法所提取纹理特征的有效性,将其用于图像检索实验。实验结果表明,新算法在检索精度上相比其他算法具有较大的提高。

In order to overcome the disadvantages of gray level co-occurrence matrix and its improved algo- rithm, such as large computational complexity, lack of computing spatial distribution information and ignoring the image correlation properties, which leads to the effective information of image texture can not be expressed well, a new texture feature used for image retrieval is proposed. Firstly, combined with the statistical character- istics and neighbor information, the relation matrix is extracted. Then, for the different scale of the neighbor- hood, the multi-neighborhood spatial distribution feature is constructed and used for image retrieval. The analy- sis shows that the texture feature extracted by this algorithm has small amount of computation and low complex- ity, and the spatial texture distribution of the image can be well described because the structural features and statistical features of the texture are effectivedly combined. The new algorithm is applied to the image retrieval experiment to prove the effectiveness of the texture feature extracted. The experimental results show that this al- gorithm has better retrieval efficiency and accuracy than other algorithms.