A Combined Algorithm for Layout Analysis of Arabic Document Images and Text Lines Extraction

Abstract

Text and not-text segmentation and text line extraction from document images are the most challenging problems of information indexing of Arabic document images such as books, technical articles, business letters and faxes in order to successfully process them in systems such as OCR. Researches on Arabic language related to documents digitization have been focusing on word and handwriting recognition. Few approaches have been proposed for layout analysis for Arabic scanned/captured documents. In this paper we present a page segmentation method that deals with the complexity of the Arabic language characteristics and fonts using the combination between two algorithms. The first method is the Run length Smoothing. The second method is the Connected Component Labeling algorithm for text and non-text classification using SVM. The combination of the two methods is based on Anding and Oring operations between the outputs of the two methods based on certain conditions. Then, dynamic horizontal projection based on dynamic updating of the threshold to commensurate with the noise associated with different documents and in between text lines. The performance evaluation is performed using manually generated ground truth representations from a dataset of Arabic document images captured using cameras and a hardware built for this purpose. Evaluation and experimental results demonstrate that the proposed text extraction method is independent from different document size, text size, font, shape, and is robust to Arabic document segmentation and text lines extraction.

A Review of Arabic Document Analysis Methods

“…Elanwar et al [16,20,23] proposed various analyses based on SVM (support vector machine) classifiers for extracting six logical labels from book pages. The same classifier was utilized by Alshameri et al [19] and Hesham et al [13,22] for text and non-text segmentation. Another learning technique used for segmentation and classification is neural network classification (Multilayer Perceptron-Back propagation) [9], whereas Ahmed et al [10] used k-means clustering and Gaussian Mixture Modelling (GMM).…”

“…Many studies successfully select the best binarization method relative to the benchmarking dataset's category. Several studies [11][12][13]22] have confirmed the efficiency of adaptive thresholding [35], such as Otsu thresholding [26] and Sauvola thresholding [27], whereas other studies [7,12,19] used filters for denoising, such as median filters [31] and Gaussian filters. The noise is presented as marginal noise, background noise, edge noise, rule line noise, pattern noise, and salt and pepper noise that can be created during scanning, transmission, or conversion to digital form, which affects the analysis process.…”

“…• The top-down approach: where the page is considered as a single large zone, and then it will be successively divided into smaller zones until no zone remains for more division, which relies on finding the largest white rectangles possible and computing the horizontal and vertical histograms of the foreground pixels. A small number of studies [7,9,16] used the XY cut algorithm [37], whereas the well-known algorithm projection profile analysis [29] was used by a significant number of studies [7,8,11,12,14,15,[17][18][19];…”

BCE-Arabic-v1 dataset

“…The dataset provided by Bukhari et al [5] contains 25 images from books and newspapers, including multi-script images that contain both English and Arabic script; the Hadjar and Ingold datasets [11,12,13] contain between 50 to 150 pages from three different newspapers (Annahar, AL Hayat, and AL Quds), and the dataset by ElShameri et al [3] contains 200 pages from newspapers. The database by the Environmental Research Institute of Michigan [26] consists of 750 images of pages from machine-printed Arabic books and magazines.…”

The application of new methods for offline recognition in printed Arabic documents

“…• Alshameri et al in [4] presented a method for text/non-text segmentation and text line extraction from document images, where they used RLSA, CCs for text segmentation and an SVM for figure detection, by applying the ANDing and ORing operations to set the correct bounding-box for each category (text/figure). This technique gave interesting results, but the application of their RLSA is efficient only in certain special cases, where specific thresholds have to be applied, and specific vertical/horizontal projections are used to distinguish between CCs with a special spatial structure.…”