Local Thresholding Algorithm Based on Variable Window Size Statistics

In an automatic document conversion system, which builds digital documents from scanned articles, there is a need to perform various adjustments before the scanned image is fed to the layout analysis system. This is because the layout detection system is sensitive to errors when the page elements are not properly identified, represented, denoised, etc. Such an adjustment is the detection of foreground and background of a document or simply called a document image binarization. This paper presents a new idea for treating the common problems which may occur during the binarization phase of the documents, that considers a parameter-free local binarization algorithm which dynamically computes the window size after it sets a threshold for the standard variation value of the window. This proved to offer more consistent results for a wide variety of scanned documents consisting of various old newspapers and old library documents in different languages, both handwritten and textual documents. Two methods for computing the local gray level threshold are proposed using the mean and standard deviation of the pixels in the current window: