基于几何流形能量最小化的图像检索算法

IMAGE RETRIEVAL ALGORITHM BASED ON GEOMETRIC MANIFOLD ENERGY MINIMISATION

提出了一种基于几何流形能量GEOMEN(geometric manifold energy)最小化的图像检索算法。许多基于流形的检索算法都是在图像的特征空间提取相应的语义流形空间,进而在语义空间中进行图像检索排序。将图像的检索看作一个图像数据库中搜索一个最优图像能量环的问题。图像能量环表示了图像之间的联系和相关性,通过最小化GEOMEN可以得到最优图像环。最小化的求解涉及到一个组合优化的问题,传统的禁忌搜索算法在选择最优候选集时非常耗时,提出一种智能的积极禁忌搜索算法求解最优环,实验表明提出的算法检索性能高,可以得到较高的查全率与准确率。

This paper proposes a novel method for image retrieval based on geometric manifold energy （GEOMEN） minimisation. Most manifold based retrieval methods extract corresponding semantic manifold space from image feature space and then do image retrieving and ranking in semantic space. In our work reported in the article, we treat image retrieval as a problem of searching for an optimal energy cycle in the image database. The cycle represents the connection and correlation between images, and the optimal image cycle can be found by min- imising the GEOMEN. To resolve minimisation refers to a combinatorial optimisation problem. Since traditional tabu search method is very time-consuming in picking up the optimal candidate set, in this article, we propose an intelligent optimisation method, named active tabu search, to find the optimised circle. Experiments show that the propose method has high retrieval performance and can achieve higher recall ratio and precision.