Synthetic data for Arabic OCR system development

Abstract

A system for the automatic generation of synthetic databases for the development or evaluation ofArabic word or text recognition systems (Arabic OCR) is presented. The proposed system works without any scanning of printed papel: Firstly Arabic text has to be typeset using a standard typesetting system. Secondly a noise-free bitmap of the document and the corresponding ground truth (GT) is automatically generated. Finally, an image distortion can be superimposed to the character or word image to simulate the expected real world noise of the intended application. All necessary modules are presented together with some examples. Special problems caused by specijic features of Arabic, such as printing from right to left, many diacritical points, variation in the height of characters, and changes in the relative position to the writing line, are suggested. The synthetic data set was used to train and test a recognition system based on Hidden Markov Model (HMM), which was originally developed for German cursive script, for Arabic printed words. Recognition results with direrent synthetic data sets are presented.

Handwriting synthesis: classifications and techniques

“…Some techniques may involve parameters which require expert knowledge for calibration while other parameters may be trained from the data available. Moreover, the number of parameters that need to be trained also places some constraint on the minimum data required to robustly train the model [54]. But sometimes, more parameters provide increased flexibility in deciding the desired quality and property of the synthesized text.…”

Cognitive Inspired Model to Generate Duplicated Static Signature Images

“…These values were randomly selected by a uniform distribution U(a, b), where (a, b) are the minimum and maximum values, as follows: A x = h x /U(20, 70) and A y = h y /U(20, 70) for the horizontal and vertical amplitude respectively and for the number of periods: N x = U(0. 1,2) and N y = U(0.1, 2). Finally, the new signature trajectory plan is obtained linking the new dots using Bresenham's line.…”

“…Generating duplicated synthetic handwriting samples of a person from a limited number of real samples has already been proposed [1], [2]. Focusing on handwriting signatures, different strategies have been used, which can be classified into two approaches:…”

Groundtruth Generation and Document Image Degradation

“…In [12], synthetic data is generated and used for training a Hidden Markov Model (HMM) based Arabic OCR system. Symbolic ground truth is keyed in and formatted in a LATEX environment, while the noise free images are obtained from the DVI files.…”