Text Enhancement in Projected Imagery

There is great interest in improving the visual quality of projected imagery. In particular, for image enhancement, we would assert that text and non-text regions should be enhanced differently in seeking to maximize perceived quality, since the spatial and statistical characteristics of text and non-text images are quite distinct. In this paper, we present a text enhancement scheme based on a novel local dynamic range statistical thresholding. Given an input image, text-like regions are obtained on the basis of computing the local statistics of regions having a high dynamic range, allowing a pixel-wise classification into text-like or background classes. The actual enhancement is obtained via class-dependent Wiener filtering, with text-like regions sharpened more than the background. Experimental results on four challenging images show that the proposed scheme offers better visual quality than projection without enhancement as well as a recent state-of-the-art enhancement method.