改进TLBO的相关反馈图像检索方法

针对当前基于进化算法的相关反馈图像检索方法无法很好地结合用户偏好信息和设置参数过多的问题,提出一种基于改进教与学优化的相关反馈图像检索方法。根据图像检索问题的特定环境,对教与学优化算法进行了一系列改进:首先,结合最近邻分类法构造适应度函数的约束条件,使之更好地反映用户偏好信息;其次,通过在教阶段将相关图像集的中心图像作为教师以及在学阶段将相关图像作为学员学习的对象,使算法快速收敛到相关图像区域;最后,结合约束处理技术Deb准则进行学员的选择操作。将该算法与目前效果优异的3种基于进化算法的相关反馈技术在两套标准图像测试集上进行对比。结果表明,所提算法相较于另外3种算法具有明显的优势,能更好地结合用户偏好信息提高图像检索性能。

Image retrieval method with relevance feedback based on improved teaching-learning-based optimization algorithm

Since the current content-based image retrieval with the relevance feedback (RF) methods based on the evolutionary algorithm could not well combine the user bias and need to set many parameters, a relevance feedback image retrieval method based on the improved teaching-learning-based optimization algorithm (ITLBO-RF) is proposed. Considering the situation of image retrieval, a series of improvements are implemented. Firstly, combining with the nearest-neighbor approach, the fitness function with constraint is proposed for better reflecting the user bias. Secondly, the center of the relevant images is regarded as the teacher in the teacher phase and the relevant image is regarded as the learning object in the learner phase, which make the algorithm converge fast to the region of relevant images. Finally, the selection operation of students based on Deb standards is conducted. ITLBO-RF is compared with three state of the art RFs based on the evolutionary algorithm on two benchmark images. The results show that ITLBO-RF has obvious advantage in comparison with other three algorithms, increases the performance of image retrieval and can better meet the user needs of image retrieval.