Text line processing for high-confidence skew detection in image documents

Skew detection and correction is an important step in automated content conversion systems, on which overall system performance is dependent. Although there are many working solutions at the present time, the search for an algorithm that can achieve good error rates in a fast running time and on different layout types is still open, so new solutions for skew detection are needed. The paper at hand presents a neighbor clustering based approach that has the classical advantages of this class of algorithms - the speed, but delivers better accuracy, comparable with that of Hough based solutions.