Color-Structured Image Search

Jingdong Wang, Xian-Sheng Hua, and Yinghai Zhao

6 July 2009

In this paper, we propose a novel interactive image ranking technique, color-structured image search, by exploiting color spatial relation information, and demonstrate it in image research results refinement. This technique enables users to draw a few color strokes, called target color structure, to indicate the intent. Beyond the conventional image retrieval methods based on query by image content that simply view the target color structure as an image and compute the similarity merely according to the spatial correspondence criterion, the proposed method represents the target color structure using a set of spatially relational colors, and evaluates the similarities between target color structure and image color structure by exploiting their spatial correspondence information, and an extra novel criterion, the consistency of spatial relation of colors implied in target color structure, and even offering a scheme to handle the uncertainty of target color structure. This similarity evaluation process can be accomplished efficiently. Moreover, a convenient interactive interface is presented to allow users to specify target color structure flexibly. Experimental results demonstrate the effectiveness and efficiency of the proposed approach.

By the time we were done, two more parties of two had come in, and both had ordered the "pork ribs", which the waitress explained was "really not ribs, but just various bones with some meat". Four people all making slurpy, yummy noises was enough to convince us to order that next time.