Techniques Of Adaptive Threshold Setting For Document Scanning Applications

Abstract

In the application of bilevel digital display and electronic print-ing, threshold setting is used to get binary images from optical scanning digitizers. Such images can be simply obtained from a raw continuous-tone image by using a fixed threshold. However, due to (1) a wide range of background colors, (2) wide variations of density in the printed infor-mation, and (3) the shading effect caused by imaging optics, adaptive threshold setting is obviously needed for obtaining high-quality displays and hard copies. This paper will review both known and newly developed adaptive threshold techniques. Simulated test patterns were used to make the comparison quantitatively. The techniques that will be re-viewed include the fixed threshold, peak and valley detector, gradient detector, fractional average context, transition-refreshed peak and valley detector, and robust two-dimensional adaptive thresholder. Among them, the last technique is found to perform very robustly in the presence of shading as well as text density variation. Specifically, both objective and subjective evaluations of pictorial results from this tech-nique have indicated that text with a differential density as low as 0.15AD is detectable with background shading variations as much as 0.30D (50%).