摘要
相似度衡量与特征提取在图像检索中具有重要的作用。基于自适应颜色特征提取的技术,提出一种新的颜色相似度衡量方法,称作加权主色优先距离WMCF。它由3个视觉感知特性条件导出,据此改善图像检索效果。同时,也采用简化的脉冲耦合神经网络提取新的纹理特征,进一步提高图像检索的精确度。实验表明,新的相似度衡量方法相较于CHIC及OCCD衡量方法有着更高的精确度和较小的时间复杂度;同时结合颜色特征与纹理特征的最终检索方法相较FC、BDIP Nmi等方法在精确度上有10%左右的提高,且具有更好的相关图像排序结果。
Similarity measurement combined with feature extraction plays an important role in image retrieval. Based on adaptive color feature extraction technique, this paper proposes a new color similarity measurement method called weighted main color first （WMCF） distance, which is derived from three conditions approximating human visual perception and used to improve the retrieval performance in CBIR. Meanwhile, a simplified pulse-coupled neural network （PCNN） is adopted to extract the new texture feature, which is combined with color feature to improve the image retrieval performance further. Experiment results show that the new similarity measurement meth- od can get better performance and much lower time complexity compared with the similarity measurement methods of histogram by cluste- ring （ CHIC ） and optimal color composition distance （OCCD）. compared with fixed cardinality （ FC ）, block difference of inverse proba- bilities （BDIP） and Normalized Moment of Inertia （Nmi） methods, the final retrieval method combining color feature and texture feature can have nearly 10% performance improvement and better ANMRR index.
出处
《仪器仪表学报》
EI
CAS
CSCD
北大核心
2014年第10期2286-2292,共7页
Chinese Journal of Scientific Instrument
基金
国家自然科学基金(61371148)
上海自然科学基金(12ZR1402500)资助项目
