A Fuzzy C-Means Based Approach Towards Efficient Document Image Binarization

Many traditional binarization techniques fail to overcome the challenging impediments fostered by degraded historical handwritten document images. In this paper, we present a fast and competent, yet simple binarization technique that uses a Fuzzy C-Means based global thresholding approach, aided by background separation. The proposed method uses a superset of foreground regions to correctly assess background of the document image. Background is estimated based on a sliding interpolation window of variable dimension, judged by appraising the nature of text stroke. Ultimately a global approach is undertaken to binarize the background-separated normalized and enhanced image by clustering the pixels using Fuzzy C-Means. This helps considering indeterministic nature of each pixel and the bland nature of the normalized image. The proposed technique is applied on the most recent (2016) benchmarking dataset of Handwritten counterpart of Document Image Binarization Contest (H-DIBCO). In order to substantiate its competence and accuracy, the experimental results are compared with the top-three winning techniques in the contest and other well-known techniques, in terms of an ensemble of parameters. It is observed that the proposed technique outperforms the best competing techniques in almost all the measuring parameters.