Amharic Character Recognition Based on Features Extracted by CNN and Auto-Encoder Models

Obsie, Efrem Yohannes; Qu, Hongchun; Qingqin, Huang

doi:10.1145/3474963.3474972

Cited by 4 publications

(2 citation statements)

References 32 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Recognizing handwritten Ethiopic text is challenging due to its vast character set, complex shape, variations in handwriting styles, incomplete strokes, and noise in scanned images. Despite research in pattern recognition for popular scripts, Ethiopic text recognition has not received comparable attention in OCR research [7], [8]. Previous works [9], [10], [11], [12] on Ethiopic script OCR have primarily been based on printed texts, and the recognition of handwritten Ethiopic scripts has remained relatively unexplored [8], [13] due to the scarcity of public research datasets [13].…”

Section: Introductionmentioning

confidence: 99%

End-to-End Historical Handwritten Ethiopic Text Recognition Using Deep Learning

Malhotra,

Addis

2023

IEEE Access

View full text Add to dashboard Cite

Recognizing handwritten text is a challenging task, especially for scripts with numerous alphabets and symbols. The Ethiopic script has a vast character set and is used for historical documents in typewritten, handwritten, and hand-printed forms. However, despite its importance as an ancient script, optical character recognition research has not given enough attention to Ethiopic text recognition. In recent years, deep learning (DL) has emerged as a powerful technique for recognizing patterns. In this study, a DL approach is used to recognize historical Ethiopic handwritten texts. The recognition model employs an endto-end strategy enabling sequential feature extraction and efficient recognition. An attention mechanism coupled with a connectionist temporal classification architecture is the core of this recognition model architecture. In addition, there are seven convolutional neural networks and two recurrent neural networks. We increase the training data using data augmentation techniques to address the data scarcity common in deep learning applications. The experiments include an original training dataset of 79,684 historical handwritten images and an augmented dataset of 10,000 images containing Ethiopic texts. The model used for recognition showed promising results. For ''Test Set I'' which had 6,150 samples, the character error rate (CER) was 17.95%, and for ''Test Set II'' which had 15,935 samples, the CER was 29.95%. These outcomes indicate that this approach has the potential to improve the recognition of historical handwritten Ethiopic text.INDEX TERMS Deep learning, end-to-end learning, ethiopic script, handwritten text recognition, pattern recognition.RUCHIKA MALHOTRA received the master's and Ph.D. degrees in software engineering from

show abstract

Section: Introductionmentioning

confidence: 99%

End-to-End Historical Handwritten Ethiopic Text Recognition Using Deep Learning

Malhotra,

Addis

2023

IEEE Access

View full text Add to dashboard Cite

show abstract

“…Even though such studies will have their own benefit by showing how universally proposed solutions will fit specific problem domains, on the other hand however will hinder innovations that could emerge because of specific problems. In this regard, while contextualized and innovative techniques to address Amharic script specifically is very important, it is overlooked in emerging research works [1,8,18].…”

Section: Introductionmentioning

confidence: 99%

Improving Amharic Handwritten Word Recognition Using Auxiliary Task

Gondere¹,

Schmidt-Thieme²,

Sharma³

et al. 2022

Preprint

View full text Add to dashboard Cite

Amharic is one of the official languages of the Federal Democratic Republic of Ethiopia. It is one of the languages that use an Ethiopic script which is derived from Gee'z, ancient and currently a liturgical language. Amharic is also one of the most widely used literature-rich languages of Ethiopia. There are very limited innovative and customized research works in Amharic optical character recognition (OCR) in general and Amharic handwritten text recognition in particular. In this study, Amharic handwritten word recognition will be investigated. State-of-theart deep learning techniques including convolutional neural networks together with recurrent neural networks and connectionist temporal classification (CTC) loss were used to make the recognition in an end-toend fashion. More importantly, an innovative way of complementing the loss function using the auxiliary task from the row-wise similarities of the Amharic alphabet was tested to show a significant recognition improvement over a baseline method. Such findings will promote innovative problem-specific solutions as well as will open insight to a generalized solution that emerges from problem-specific domains.

show abstract