A Fast Corner Detection Algorithm Based on Area Deviation

A new algorithm, based on area-deviation, is proposed for the detection of corner points of digitized curves. The algorithm consists of two steps. In the first step, a fixed-length chord is moving along the digitized curve step by step, and the area between the chord and the associated curve segment is measured. The area values are used to represent the average curvature values for the corresponding curve segments. The curve segments with their area values having reached a local maximum and exceeded a threshold value will be identified. Each such curve segment will contain one corner point. In the second step, the exact position of the comer point in each identified curve segment is found by comparing the changes in the curve's directions at each point of the curve segment. A lot of graphical objects including Chinese and English fonts have been tested. The algorithm is proved to be efficient and gives results close to human expectations.