Data Augmentation for Offline Arabic Handwritten Text Recognition Using Moving Least Squares
Bibliographic record
Abstract
This paper addresses the research problem of Offline Arabic Handwriting Text Recognition (HTR).One of the most important approaches to HTR systems is deep learning.A large amount of annotated data is needed to train deep learning-based HTR systems.The Arabic language is spoken by hundreds of millions of people in North Africa and the Middle East.Writing styles and common words differ significantly between those regions.Due to the great diversity possible, designing a statistically represented and balanced database of Arabic handwritten texts by gathering and labeling the texts is an arduous task to achieve.One of the ways to enrich the training databases is by augmenting the existing data.We have developed a new data augmentation technique for Arabic handwritten texts using Moving Least Squares (MLS) to deform the images.This technique results in realistic images that look like manipulating real-world images, and the deformations are done using linear functions that produce deformations in real time.We aim to deform the training data images randomly in a way that the text present in the images is still recognizable by a human.This augmentation technique can be used directly on images to augment them unlike other techniques such as Generative Adversarial Networks (GAN) where they must be trained beforehand.At the same time, it produces new complex augmented images compared to simple traditional augmentation techniques such as rotations and translations.In addition to this augmentation technique, we used a deep learning system called Convolutional Recurrent Neural Networks (CRNN) to test the new technique, and we have experimented with a CRNN model that accepts small input-size images to boost the time needed for both training and image augmentations.All the experimentations are carried out on the Arabic IFN/ENIT database.The results show that the small input size CRNN model outperforms the large input size CRNN model by a big margin.The results also show that the integration of images augmented by the MLS technique can help the recognition system to generalize better on the test data, therefore, it can slightly improve the performance of the recognition system.
Fetched live from OpenAlex and de-inverted. Abstracts are not stored in this database: the inverted indexes are 8.6 GB of the frame’s 9.3 GB of text, and the host has 13 GB free.
How this classification was reachedexpand
Full frame distilled prediction
Teacher imitationNot calibrated prevalence, not ground truth. Human validation pending. Learned from the 10,348 direct Codex labels and 10,348 direct Gemma labels. Candidate is the union of thresholded teacher heads; consensus is their intersection. These outputs are machine_predicted_unvalidated and are not human labels or direct frontier model labels.
Codex and Gemma teacher scores by category
| Category | Codex | Gemma |
|---|---|---|
| Metaresearch | 0.001 | 0.000 |
| Meta-epidemiology (narrow) | 0.000 | 0.000 |
| Meta-epidemiology (broad) | 0.000 | 0.000 |
| Bibliometrics | 0.000 | 0.001 |
| Science and technology studies | 0.000 | 0.000 |
| Scholarly communication | 0.001 | 0.001 |
| Open science | 0.001 | 0.000 |
| Research integrity | 0.000 | 0.000 |
| Insufficient payload (model declined to judge) | 0.000 | 0.000 |
Machine scores (provisional)
The two teacher heads of the student model, read on this work. A score orders the frame for review; it never asserts a category, and the validation status ships verbatim with every row.
Baseline scores from an immature model (maturity gate not passed, 7 training rounds). Scores rank; they never assert a category.
score_only:v0-immature-baseline · verbatim from the scoring run: score_only means the number may rank works, and no category label ships from itClassification
machine, unvalidatedMachine predicted; a candidate call from one teacher head, not a consensus.
How this classification was reached, model by model and score by score, is at the end of the page under "How this classification was reached".