基于梯度相关图的光照不变图像检索算法  

ILLUMINATION INVARIANT IMAGE RETRIEVAL ALGORITHM BASED ON GRADIENT CORRELOGRAM

作　　者：王小龙[1] 沈新宁 杜建洪[1] 

机构地区：[1]复旦大学信息科学与工程学院,上海200433

出　　处：《计算机应用与软件》2014年第6期127-131,共5页Computer Applications and Software

摘　　要：针对现有图像光照不变特征提取算法存在特征矢量尺寸较大,且对于场景的尺度、平移及视角等变化鲁棒性较差的缺点,提出将相关图算法与光照不变导m1m2m3相结合的基于梯度相关图的特征提取算法。算法首先计算图像的光照不变导,去除光照对图像的影响,然后提取光照不变导梯度图像的相关图特征作为特征索引。实验结果表明该梯度相关图算法在存在光照影响时对图像的检索性能优于以往常用算法,且对于场景的尺度、平移、视角等变化具有很好的鲁棒性。Existing feature extraction algorithms for the images with invariant illumination have the problems of large size in eigenvector and poor robustness in variations of scaling, translation and visual angles of scenes. In light of this, we propose the gradient correlogrambased feature extraction algorithm, which combines the correlogram algorithm with the illumination invariant derivatives m1m2m3. The algorithm first calculates the illumination invariant derivatives of the image to remove the interference of illumination on the image. Then it extracts the correlogram feature of the gradient image of illumination invariant derivatives as the feature indices. Experimental result shows that, under the influence of illumination, the presented gradient correlogram algorithm has better image retrieval performance than the previous commonly used algorithms, and is robust to the changes in scaling, translation and visual angles of the scene.

关 键 词：光照不变导 颜色不变量 梯度相关图 图像检索 特征提取鲁棒性 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]