Applying Preattentive Visual Guidance in Document Image Analysis

In this paper, we present a novel methodology on document image analysis (DIA) which harnesses the mechanism of preattentive visual guidance in human vision. Summarizing the psychophysical research on preattentive vision, we suggest using two types of computations to simulate this biological process: the visual similarity clustering and visual saliency detection. Based on the computational implementation of these two processes, we develop a biological plausible way to guide the interpretation of document images, which is distinctively different from previous DIA methods. Experimental results prove the effectiveness of these two processes, whose outputs can be further utilized by any task-oriented DIA application.