Improvement of binarization performance by applying DCT as pre-processing procedure

Abstract

This paper proposes a new pre-processing approach to a sequence of binarization process, by applying the discrete cosine transform (DCT) to document image prior to binarization. This procedure can be used as a complement or alternative to denoising process used in many binarization methods. Firstly, original document images are transformed by the DCT. Then, the transformed document images are passed to standard image binarization methods, namely Otsu, Niblack, Sauvola and NICK. Secondly, original document images are directly binarized by the same methods. Then performance of the latter are compared with that of the former in terms of recall and precision. Performance evaluation are conducted by using noiseless and noisy ancient Indonesian documents with Arabic characters. It turns out that this pre-processing procedure improves binarization performance, especially for noisy documents.