An Improved Adaptive Document Image Binarization Method

Dokumen Jawi kuno merupakan warisan budaya yang berisi informasi penting tentang peradaban masa lalu yang dapat dijadikan pedoman untuk masa sekarang ini. Dokumen Jawi kuno telah mengalami penurunan kualitas yang disebabkan oleh beberapa faktor seperti kualitas kertas atau karena proses penyimpanan. Penurunan kualitas ini menyebabkan informasi yang terdapat pada dokumen tersebut menghilang dan sulit untuk diakses. Artikel ini mengusulkan metode binerisasi untuk membangkitkan kembali informasi yang terdapat pada dokumen Jawi kuno. Metode usulan merupakan kombinasi antara metode binerisasi berbasis nilai ambang lokal dan global. Metode usulan diuji terhadap dokumen Jawi kuno dan dokumen uji standar yang dikenal dengan nama Handwritten Document Image Binarization Contest (HDIBCO) 2016. Citra hasil binerisasi dievaluasi menggunakan metode: F-measure, pseudo F-measure, peak signal-to-noise ratio, distance reciprocal distortion, dan misclasification penalty metric. Secara rata-rata, nilai evaluasi F-measure dari metode usulan mencapai 88,18 dan 89,04 masing-masing untuk dataset Jawi dan HDIBCO-2016. Hasil ini lebih baik dari metode pembanding yang menunjukkan bahwa metode usulan berhasil meningkatkan kinerja metode binerisasi untuk dataset Jawi dan HDIBCO-2016. AbstractAncient Jawi document is a cultural heritage, which contains knowledge of past civilization for developing a better future. Ancient Jawi document suffers from severe degradation due to some factors such as paper quality or poor retention process. The degradation reduces information on the document and thus the information is difficult to access. This paper proposed a binarization method for restoring the information from degraded ancient Jawi document. The proposed method combined a local and global thresholding method for extracting the text from the background. The experiment was conducted on ancient Jawi document and Handwritten Document Image Binarization Contest (HDIBCO) 2016 datasets. The result was evaluated using F-measure, pseudo F-measure, peak signal-to-noise ratio, distance reciprocal distortion, dan misclassification penalty metric. The average result showed that the proposed method achieved 88.18 and 89.04 of F-measure, for Jawi and HDIBCO-2016, respectively. The proposed method resulted in better performance compared with several benchmarking methods. It can be concluded that the proposed method succeeded to enhance binarization performance.

Download Full-text

U-Net-bin: hacking the document image binarization contest

Computer Optics ◽

10.18287/2412-6179-2019-43-5-825-832 ◽

2019 ◽

Vol 43 (5) ◽

pp. 825-832 ◽

Cited By ~ 2

Author(s):

P.V. Bezmaternykh ◽

D.A. Ilin ◽

D.P. Nikolaev

Keyword(s):

Ground Truth ◽

Document Image ◽

Training Data ◽

Data Preparation ◽

Historical Document ◽

Image Binarization ◽

Biomedical Image ◽

Document Image Binarization ◽

Binarization Method ◽

Document Binarization

Image binarization is still a challenging task in a variety of applications. In particular, Document Image Binarization Contest (DIBCO) is organized regularly to track the state-of-the-art techniques for the historical document binarization. In this work we present a binarization method that was ranked first in the DIBCO`17 contest. It is a convolutional neural network (CNN) based method which uses U-Net architecture, originally designed for biomedical image segmentation. We describe our approach to training data preparation and contest ground truth examination and provide multiple insights on its construction (so called hacking). It led to more accurate historical document binarization problem statement with respect to the challenges one could face in the open access datasets. A docker container with the final network along with all the supplementary data we used in the training process has been published on Github.

Download Full-text