Binarization of Colored Document Images using Spectral Clustering
Enas M. Elgbbas, Mahmoud I. Khalil, Hazem M. Abbas · 2018
In this paper, we propose a hybrid method for text binarization of historical documents. Proposed method incorporates the advantages of Otsu and spectral clustering algorithm. In text binarization problem, there are noise and faint text challenges. To overcome the noise problem, a preprocessing step is applied to the colored document. After that, the resulted image is binarized using Otsu, producing a global binary image. As a final step, the spectral clustering algorithm is locally applied to the original image, with the aid of the global binary image, in order to retrieve faint text. The proposed design of spectral clustering provides a significant reduction of the similarity matrix computing time and size use, without affecting the quality of clustering. This method is suitable for colored document images. The efficiency of this method is illustrated by experiments using DIBCO images.