Document Image Binarization

Principal stage of the document image analysis procedure is the binarization, according to which the pixels are classified into text and background. It is a crucial stage that can affect further stages including the final character recognition stage. This thesis is focused on document image binarization, including both binarization techniques and evaluation methodologies. Specifically, according to the developed performance evaluation methodologies, the pixel-level ground-truth image is constructed using a semi-automatic procedure based on the edges and the skeleton of the characters. The new measures use (a) weights that start from the ground truth contour and (b) the local stroke width to limit the weights close to the character areas and to properly normalize those weights. Experimental results prove the validity and effectiveness of the new measures for document images, while other measures concern the image or signal processing area in general. Concerning binarization techniques, some improvements were initially proposed for the well-known technique of Yang&Yan. To further enhance the quality of binarization and be more robust against different types of degradations (e.g. faint characters, bleed-through and non-uniform background), a new binarization technique was developed that was based on background estimation and on the combination of selected global and local binarization techniques. Additionally, a binarization technique was developed for the binarization of the text areas captured from video content. This technique is also based on the Yang&Yan binarization technique and sets low and high values in its global parameter for the inside and outside area of the text. Initially, the definition of the text areas is based on the baselines of the text and at the final stage the text areas are better defined by the convex hulls of neighbouring textual components. Furthermore, through the document image binarization contests that we organized, a publicly available benchmark has been created that aids in the development of document image binarization techniques and evaluation methodologies.