Retrieving images using saliency detection and graph matching

摘要: The need for fast retrieving images has recently increased tremendously in many application areas (biomedicine, military, commerce, education, etc.). In this work, we exploit the saliency detection to select a group of salient regions and utilize an undirected graph model dependency among these regions, so that similarity can be measured by calculating corresponding graphs. Identification pixels decrease interferences from irrelevant information, make image representation more effective. introduction better characterize spatial constraints regions. comparison experiments are carried out on three representative datasets publicly available (Holidays, UKB, Oxford 5k), experimental results show integration proposed method SIFT-like local descriptors improve existing state-of-the-art retrieval accuracy.