Optical Character Recognition Recognition Research Articles

Data Testing in Digital Transmission: Tamil Literature

The developing area of Optical Character Recognition (OCR) is digital handwriting recognition. Manual writing is replaced by a digital writing pad. The font and shape of the letters vary while writing digitally. The writer’s digital pen pressure and position on the digital pad cause covert text file problems during OCR recognition. When converting OCR to text, an error occurs because of the variations in letter shapes. In languages like Tamil, Chinese, Arabic, and Telugu, where the alphabet is made up of bends, curves, and rings, the aforementioned issue occurs. Tamil has more word mistakes in OCR-to-text conversion because the alphabet comprises curves and angles that must be correctly transcribed. The ResNet (Residual Neural Network) Two-Stage Bottleneck Architecture (RTSBA) is suggested in this paper. In order to recognize text written in Tamil on a digital writing pad, this article suggests using ResNet (Residual Neural Network) Two-Stage Bottleneck Architecture (RTSBA). The suggested RTSBA reduces the complexity of the Tamil alphabet recognition problem by using two distinct phases of neural networks. There are fewer inputs and variables in the early stages. Time and computational complexity are minimized in the last phase. A two-channel and two-stream transformer, long short-term memory, Inception-v3, recurrent neural networks, convolutional neural networks, and other conventional algorithms have been compared to the suggested algorithm. The digital writing pad-handwritten and HP lab datasets demonstrate the effectiveness of proposed methods like RTSBA, which yield accuracy rates of 98.7% and 97.1%, respectively.

Tamil OCR Conversion from Digital Writing Pad Recognition Accuracy Improves through Modified Deep Learning Architectures

Digital handwritten recognition is an emerging field in optical character recognition (OCR). A digital writing pad replaces manual writing. In digital writing, the alphabet changes in font and shape. During OCR recognition, covert text file errors occur due to digital pen pressure and digital pen position on the digital pad by the writer. The shape changes in the alphabet lead to an error during the conversion of OCR to text. The above problem arises in Tamil, Chinese, Arabic, and Telugu, where the alphabet consists of bends, curves, and rings. OCR‐to‐text conversion for the Tamil language has more word errors due to angles and curves in the alphabet, which need to be converted accurately. This paper proposes ResNet two‐stage bottleneck architecture (RTSBA) for Tamil language‐based text recognition written on a digital writing pad. In the proposed RTSBA, two separate stages of neural networks reduce the complexity of the Tamil alphabet recognition problem. In the initial stage, the number of inputs and variables is reduced. In the final stage, time and computation complexity are reduced. The proposed algorithm has been compared with traditional algorithms such as long short‐term memory, Inception‐v3, recurrent neural networks, convolutional neural networks, and a two‐channel and two‐stream transformer. Proposed methods, such as RTSBA applied in the digital writing pad‐handwritten and HP lab datasets, achieved an accuracy of 98.7% and 97.1%, respectively.

A Smart Visual Sensing Concept Involving Deep Learning for a Robust Optical Character Recognition under Hard Real-World Conditions.

In this study, we propose a new model for optical character recognition (OCR) based on both CNNs (convolutional neural networks) and RNNs (recurrent neural networks). The distortions affecting the document image can take different forms, such as blur (focus blur, motion blur, etc.), shadow, bad contrast, etc. Document-image distortions significantly decrease the performance of OCR systems, to the extent that they reach a performance close to zero. Therefore, a robust OCR model that performs robustly even under hard (distortion) conditions is still sorely needed. However, our comprehensive study in this paper shows that various related works can somewhat improve their respective OCR recognition performance of degraded document images (e.g., captured by smartphone cameras under different conditions and, thus, distorted by shadows, contrast, blur, etc.), but it is worth underscoring, that improved recognition is neither sufficient nor always satisfactory—especially in very harsh conditions. Therefore, in this paper, we suggest and develop a much better and fully different approach and model architecture, which significantly outperforms the aforementioned previous related works. Furthermore, a new dataset was gathered to show a series of different and well-representative real-world scenarios of hard distortion conditions. The new OCR model suggested performs in such a way that even document images (even from the hardest conditions) that were previously not recognizable by other OCR systems can be fully recognized with up to 97.5% accuracy/precision by our new deep-learning-based OCR model.

Removing light interference to improve character recognition rate by using single-pixel imaging

Optical Character Recognition (OCR), which is generally based on figure recognition, is mainly effective for extracting text from pictures. However, it is a challenge to maintain a high recognition rate without being affected by light interference in a complicated environment. To reduce the light interference for improving the OCR recognition rate, this article proposes a method utilizing single-pixel imaging to obtain images without light interference for character recognition. Unlike the traditional CCD/CMOS-based imaging relying on image post-processing to eliminate interference, the method proposed without the need of any image post-processing which is a cumbersome process and often leads to a deterioration in the quality of the obtained images. Theoretical analysis shows that single-pixel imaging can eliminate the light interference in certain conditions completely. The verification experiments demonstrate that single-pixel imaging can only image the target and ignore an introduced highlighted stripe, whereas traditional CCD/CMOS-based imaging can not eliminate the effects of the stripe without image post-processing. Comparison experiments in contrast with the character recognition rate of the picture obtained from traditional imaging demonstrate the counterpart obtained from single-pixel imaging increased from 88.64% to 97.73%. The evaluation index of correlation and image quality illustrates that the picture obtained by single-pixel imaging has a higher correlation with the target picture and better image quality than the counterpart of the picture obtained from traditional imaging. When the character recognition rate of the picture obtained by traditional imaging is 0% by adding a serious interference pattern from a projector, the counterpart by single-pixel imaging is 95.45%. This approach provides a welcome boost to the development of the application of single-pixel imaging in character recognition by reducing the light interference, which provides a potential direction for the development of OCR to obtain higher recognition success rate under a complex environment.

Speed Breakers, Road Marking Detection and Recognition Using Image Processing Techniques

This paper presents a image processing technique for speed breaker, road marking detection and recognition. An Optical Character Recognition (OCR) algorithm was used to recognize traffic signs such as “STOP” markings and a Hough transform was used to detect line markings which serves as a pre-processing stage to determine when the proposed technique does OCR or speed breaker recognition. The stopline inclusion serves as a pre-processing stage that tells the system when to perform stop marking recognition or speed breaker recognition. Image processing techniques was used for the processing of features from the images. Local Binary Pattern (LBP) was extracted as features and employed to train the Support Vector Machine (SVM) classifier for speed breaker recognition. Experimental results shows 79%, 100% “STOP” sign and speed breaker recognitions respectively. The proposed system goes very well for the roads which are constructed with proper painting irrespective of their dimension.

A CAPTCHA model based on visual psychophysics: Using the brain to distinguish between human users and automated computer bots

Demand for the use of online services such as free emails, social networks, and online polling is increasing at an exponential rate. Due to this, online service providers and retailers feel pressurised to satisfy the multitude of end-user expectations. Meanwhile, automated computer robots (known as “bots”) are targeting online retailers and service providers by acting as human users and providing false information in order to abuse their service provisioning. CAPTCHA is a set of challenge/response protocol, which was introduced to protect online retailers and service providers from misuse and automated computer attacks. Text-based CAPTCHAs are the most popular form, and are used by most online service providers to differentiate between the human users and bots. However, the vast majority of text-based CAPTCHAs have been broken using the Optical Character Recognition (OCR) techniques and thus, reinforces the need for developing a secure and robust CAPTCHA model. Security and usability are the two fundamental issues that pose a trade-off in the design of a CAPTCHA; a hard CAPTCHA model could also be difficult for human users to resolve, which affects its usability, and vice versa. The model developed in this study uses the unsurpassed abilities of the Human Visual System (HVS) to superimpose and integrate complex information presented in individual frames, using the mechanism of trans-saccadic memory. In this context, the model integrates in its design the concept of persistence of vision, which enables humans to see the world in a continuous fashion. Preliminary results from the proposed model based on this technique are encouraging. To ensure the usability of the proposed CAPTCHA model, we set the threshold for the ORO parameter at 40%. This ensured that our CAPTCHA strings would be recognised by human observers at a rate of over 99% (or as close to 100% as is realistic). In turn, when examining the robustness of our VICAP model to computer programme attacks, we can observe that for the traditional case of OCR recognition, based on a single-frame scenario, the Computer Recognition Success Rate (CRSR) was about 0%, while in the case of a multi-frame scenario, the CRSR could increase to up to 50%.

Text Retrieval from Natural and Scanned Images

Digital documents are easy to handle, share and store than hard copy of documents. These made people to prefer digital document over hard copy of documents. Digital documents are nothing but scanned images of a document or natural images of notice boards, traffic signs. Text detection is an important process required to extract text from images. Text from images can be extracted using Optical Character Recognition (OCR). OCR works in three phases as preprocessing, segmentation, character recognition. Preprocessing is the first phase which uses different techniques for making text easy to extract from images. In segmentation phase, each character is isolated. Then this will be given as input to OCR recognition phase which will compare it with training data-set and will recognize character. In this survey paper, different techniques for OCR are discussed.

Graph cut-based binarisation of noisy check images

Binarisation of document images with poor contrast, strong noise, complex patterns and variable modalities in the grey-scale histograms is a challenging problem. This study proposes an algorithm for the binarisation of noisy check images to extract handwriting text using normalised graph cuts (GCs). The proposed algorithm uses a normalised GC measure as a thresholding principle to distinguish the handwriting characters from the noisy background. The authors propose a factor to penalise extracting objects that do not have the elongated shape of the characters. Improving the structural quality of the characters' skeleton facilitates better feature extraction and classification, which improves the overall performance of optical character recognition (OCR). Experimental results performed on 560 check images showed significant improvements in OCR recognition rates compared to other well-established segmentation algorithms.

Iterative Cross Section Sequence Graph for Handwritten Character Segmentation

The iterative cross section sequence graph (ICSSG) is an algorithm for handwritten character segmentation. It expands the cross section sequence graph concept by applying it iteratively at equally spaced thresholds. The iterative thresholding reduces the effect of information loss associated with image binarization. ICSSG preserves the characters' skeletal structure by preventing the interference of pixels that causes flooding of adjacent characters' segments. Improving the structural quality of the characters' skeleton facilitates better feature extraction and classification, which improves the overall performance of optical character recognition (OCR). Experimental results showed significant improvements in OCR recognition rates compared to other well-established segmentation algorithms.

Capturing Tables of Contents

Both photocopied tables of contents and originals were scanned by optical character recognition (OCR). (The threshold for acceptable OCR conversion is a break-even point of 94% accuracy; it is cheaper to rekey data than fix corrupted data if rates fall below this point.) The results suggest that OCR recognition rates are likely to support improved capture productivity for as many as 75% of tables of contents. More than half (55%) of the sample can be processed effectively.