摘要
提出了一种基于色彩量化及索引的图像检索新方法。结合人眼视觉感知特性,首先将图像划分成4×4大小的非重叠子块,通过块梯度的大小,在亮度空间将图像子块划分为视觉均衡块和非均衡块。若为均衡块,则用该块RGB空间的颜色均值作为其代表颜色值,然后转换到HSV空间,并量化成32个等级,形成32维索引直方图(S-HIST);若非均衡块,根据保持颜色矩不变技术将该图像子块在RGB空间量化两种颜色,然后转化到HSV空间,并将每种颜色量化成32个等级,形成496维的索引直方图(D-HIST)。最后,综合索引特征,进行图像检索。实验结果表明:该算法是非常有效的。
A novel image retrieval algorithm based on color quantization and indexing is proposed, which is integrated with the human vision. Firstly, the image is divided into 4 × 4 non-overlapping blocks. Each block, based on its block gradient magnitude,is classified into visual uniform or visual non-uniform block. If the block belongs to a uniform block,the average of each color component for the pixels of it in RGB space is used as its representative color,tben it is converted from RGB space into HSV space,and quantized into 32 levels,.so an indexing histogram （S_HIST）of 32 dimensions is constructed;If the block belongs to a non-uniform block,then it is quantized into two representative colors by color moment-preserving. The two colors are transformed to HSV space,and each color is quantized into 32 levels ,so an indexing histogram （D_HIST） of 496 dimensions is constructed. Finally, a novel image retrieval algorithm is formed by integrating all indexing features. The result of retrieval shows that the algorithm is effective.
出处
《光电子．激光》
EI
CAS
CSCD
北大核心
2008年第2期253-257,共5页
Journal of Optoelectronics·Laser
关键词
色彩量化
块梯度
均衡块
保持颜色矩不变
color quantization
block gradient
uniform block
color moment-preserving
