Robust character extraction across varying conditions based on local distribution
Adrianto Tedjokusumo, Heon Do Jeong, Sang-Woong Lee · 2013
The result of binarization as the main part of characters extraction is dependent in many factors including non-uniform illumination and the clarity of printed characters in an object. Especially in non-rigid object, it has its typical characteristics that can lead to poor binarization result. This paper presents a robust algorithm for character extraction of printed characters in non-rigid object. The proposed method is a modified local thresholding method which is strengthened by our two directions standard deviation algorithm in separating the characters as the foreground from its background. Modified local thresholding method selects the best threshold value of each pixel by using information of local maxima and minima of intensity value in the local window along with optimization stage which is used to increase the efficiency of computational cost. The purpose of two directions standard deviation algorithm is to segment the character's area and determine character's position which improves effectiveness of our adaptive thresholding. After extensive experiments, the proposed binarization algorithm demonstrates superior performance in extracting characters among the non-rigid object with various conditions.