摘要
针对视觉词汇树模型采用的SIFT特征只采用了灰度信息,而忽略了图像的颜色特征,而多数改进采用彩色SIFT特征又耗时巨大的问题,提出一种将颜色信息作为图像全局权值融入到视觉单词匹配模型中的算法.提取图像的颜色特征,量化并计算颜色相似度,按照颜色差异由小到大排序,根据排序结果建立颜色特征权重.将图像SIFT特征分层量化为视觉单词,为了节约大规模数据库下的检索时间,引入感知哈希原理,将图像的视觉单词映射为哈希值,这样每幅图像都转化成一串哈希序列,计算图像哈希序列之间的汉明距离,将颜色特征权重作为汉明距离的加权系数,计算图像的最终相似度.实验结果表明,本文算法对于提高检索的准确率效果显著.
Aiming at SIFT feature used by visual vocabulary tree only using gray scale information, ignoring the color feature of the image and time consuming huge when most improved methods using color-SIFT feature, this paper proposed an algorithm which put the color information as global weight of the image into the visual words matching model. This method firstly extracts color feature of the image, quantifies and calculates color similarity, builds color feature weights according to the ascending order by color difference. Then the SIFT features of image are stratified quantifies as visual words. In order to save the search time in the large-scale databases, this method brings into perceptual hash algorithm, mapping the visual words of image into the hash value. And then each image is convetted into a string of hash sequence. Lastly the method calculates Hamming distance between the image hash sequences, the color feature weights as the weighting coefficient of Hamming distance and calculating the final similarity between images. The experimental results show that the algorithm can effectively improve the accuracy of the image retrieval.
出处
《小型微型计算机系统》
CSCD
北大核心
2017年第7期1653-1657,共5页
Journal of Chinese Computer Systems
基金
国家自然科学基金项目(61104213
61573168)资助
江苏省产学研前瞻性联合研究项目(BY2015019-15)资助
关键词
颜色特征
词汇树
感知哈希
图像检索
color feature
visual vocabulary tree
perceptual hash
image retrieval
