Abstract: Color is an important visual feature. Color Correlogram(CC) algorithm is commonly used in the color based image retrieval as a feature descriptor, but most of the existing methods based on CC have problems of high computational complexity and low retrieval accuracy. Aiming at this problem, this paper proposes an image retrieval algorithm based on color autocorrelogram and mutual information. It presents a novel color feature descriptor, namely Color Mutual Information(CMI). The new color feature vector which describes the global and spatial distribution relation among different colors is obtained by calculating the average mutual information between one color and all the colors around it in the CC feature matrix, thus reducing the computational complexity. Inter-feature normalization is applied in the combination of CMI and color autocorrelogram to enhance the retrieval accuracy. Experimental result shows that this integrated method can reduce the computational complexity, improves real-time response speed and retrieval accuracy.

Key words: image retrieval, color feature, Color Correlogram(CC), Color Mutual Information(CMI), feature normalization

摘要： 颜色特征是重要的图像视觉特征，颜色相关图则是当前基于内容的图像检索中常用的特征描述符，但现有基于颜色相关图的图像检索算法存在计算复杂度高、检索精确度低的问题。为此，提出基于颜色自相关图和互信息的图像检索算法。给出一种新的颜色特征描述符——颜色互信息，通过计算颜色相关图特征矩阵中每个颜色与其周围颜色的平均互信息，得到不同颜色之间的全局及空间分布特性，并作为新的颜色特征矢量，以降低计算复杂度。同时采用外部特征矢量归一化方法结合颜色互信息与颜色自相关算法，以提高检索精确度。实验结果表明，该算法可有效降低计算复杂度，提高实时响应性能和检索精度。

关键词: 图像检索, 颜色特征, 颜色相关图, 颜色互信息, 特征归一化

CLC Number: 

• TP911.73