基于混合马尔科夫链模型和扩散的图像检索

IMAGE RETRIEVAL BASED ON MIXED MARKOV CHAIN MODEL AND DIFFUSION

提出一种简单而有效的基于图和多特征融合的通用图像检索框架。对于每个特征,给定查询对象和初始检索的图像,构造一个易处理图,其结点表示图像,边是图像之间的组对相似性得分。基于多个图的随机路径选择模型,采用混合马尔科夫链模型和多特征融合,将多个易处理图融合成融合图,并将扩散应用于融合图以减少噪声。通过与现有方法的实验对比,证明该方法的有效性,提升了基准性能。通过实验评估了该方法的组件、参数和特征组合,验证了该方法的合理性和对参数变化的鲁棒性。

This paper present a simple and effective framework for general image retrieval based on graph and multifeature fusion. For each feature,given a query object and an initially retrieved image,an easy-to-handle graph was constructed whose nodes represented the image and edges were the group-pair similarity scores between the images.Based on a multi-graph stochastic path selection model,a hybrid Markov chain model and multi-feature fusion were used to fuse multiple easy-to-handle graphs into a fusion graph and applied diffusion to the fusion graph to reduce noise. By comparing with the existing methods,it proved the validity of this method and improved the benchmark performance. The components,parameters and feature combinations of this method were evaluated through experiments,which verified the rationality of the method and the robustness to the parameter changes.