Research On Large Scale Image Retrieval Based On Semantic Segmentation

Large scale similarity search has been widely applied in the fields of computer vision and multimedia.The storage and computational efficiency are the key indicators during the search of the useful information in the massive data.Hashing based search methods are proposed to deal with the above issue,which can map the similar images into the similar compact binary codes by preserving the semantic and feature similarity.Furthermore,hashing based method can effectively improve the search performance by reducing storage and computational cost.Hence,hashing method has been the key technology of image retrieval.In the research of multi-label multi-instance application,each image can be represented by multiple instances and associated with multiple labels.Furthermore,the labels are assigned into each image not instances.In order to deal with the above difficulties,we propose the semantic segmentation based hashing methods for image retrieval.First,the deep learning technology is utilized to implement the segmentation and extract features from each image.Then,the optimal distance metric are learned to ensure the semantic consistence and positive label correlation by minimizing the objective function.Finally,the anchor hashing method is employed to yield the similar binary codes to improve the accuracy of the image retrieval.Multiple image retrieval experiments are conducted in the MSRC v2 and PASCAL VOC2012 dataset.The comparison of several state-of-the-art hashing methods further validates the superiority and feasibility of our proposed method.