基于显著区域和相关反馈的图像检索算法

Image Retrieval Algorithm Based on Salient Regions and Relevance Feedback

针对基于内容的图像检索中遇到的效率低下和语义鸿沟问题,设计并实现了一个交互式的图像检索系统。系统首先结合人类视觉注意机制提取图像显著区域,再对不同的区域进行不同特征或不同权重的描述。最后,在初次检索后应用支持向量机(Support Vector Machine,SVM)和粒子群优化(Particle Swarm Optimization,PSO)算法进行相关反馈(Rele vance Feedback,RF),使检索结果更符合用户目的。实验表明,用SVM进行反馈检索效率有大幅度提高,而PSO在小样本指导下,表现出高效的学习和快速的收敛优势。

Taking into account the low-efficiency of content-based image retrieval and the semantic gap between low-level im age features and high-level semantics,the paper designs and realizes the interactive image retrieval system.The system first uses human visual attention mechanisms to extract the salient regions of the image,and then gives different regions a variety of descrip tions depending on their different weights and features.Finally,the paper applies support vector machine(SVM) and particle swarm optimization(PSO) algorithms for relevance feedback after the first retrieval.Experimental results show that SVM substan tially improves retrieval efficiency after the first feedback,and PSO displays the advantage of effective learning and fast conver gence in the case of a small number of positive images.