China 
Africa 
Explore chapters and articles related to this topic
Review on Optical Character Recognition-Based Applications of Industrial IoT
Published in Sudan Jha, Usman Tariq, Gyanendra Prasad Joshi, Vijender Kumar Solanki, Industrial Internet of Things, 2022
Handwriting recognition, in general, is classified into two types: offline and online handwriting recognition methods. OCR is an offline process, but nowadays, the online process is also available using APIs. In offline handwriting recognition, automatic conversion of text into letter codes can be used by computer applications. The advantages of offline OCR are that it can be done years after the document is written which makes it more usable in scanning books in the library, postal codes, band checks but cannot be used in real-life applications like self-driving cars to read signboards. Different handwriting styles make it more challenging to implement recognition techniques to get better accuracy. Generally, offline recognition methods involve acquisition, segmentation, recognition, and postprocessing and output of one stage is input in the next step. But, in the online system, the two-dimensional coordinates of successive points are represented as a function of time and the order of strokes made by the writer are also available [3].
An Introduction to Handwritten Character and Word Recognition
Published in Lakhmi C. Jain, Beatrice Lazzerini, KNOWLEDGE-BASED INTELLIGENT TECHNIQUES in CHARACTER RECOGNITION, 2020
Handwriting recognition systems acquire data by means of either offline devices (such as scanners and cameras) or on-line devices (such as graphic tablets). In off-line recognition systems, the input image is converted into a bit pattern. Then, specific preprocessing algorithms prepare the acquired data for subsequent processing by eliminating noise and errors caused by the acquisition process. In contrast, on-line recognition systems use dynamic writing information, namely, the number, order, and direction of the strokes to recognize each character as it is being written. Typically, the two-dimensional coordinates of the pen position on the writing surface are sampled at fixed time intervals.
Evolution of Long Short-Term Memory (LSTM) in Air Pollution Forecasting
Published in Monika Mangla, Subhash K. Shinde, Vaishali Mehta, Nonita Sharma, Sachi Nandan Mohanty, Handbook of Research on Machine Learning, 2022
Satheesh Abimannan, Deepak Kochhar, Yue-Shan Chang, K. Thirunavukkarasu
Handwriting recognition is one of the hardest problems to accomplish as it highly depends on contextual information and lexicon matching. We, humans, can recognize and differentiate handwritings comparatively easily because we can effectively map the context knitted within it. We can differentiate similar-looking characters as well. Although, there’s a standard way to write each and every character, handwritings highly depend on individuals. Every individual’s handwriting differs on various parameters such as stroke, curves, etc. (Figure 15.2). Sometimes, we can also guess words even if the writing is not clearly understood. We now understand how hard it becomes to even translate the necessary information for handwriting recognition into recognizable data patterns. The variety available is overwhelming. Over the years, multiple approaches have been developed to solve this problem, and LSTMs perform far better than any other algorithm. Multiple researchers have confirmed the accuracy of LSTMs in handwriting recognition. This is due to the fact that handwriting can also be represented as sequential data patterns. LSTMs learn relations between adjacent characters, curves, etc., by converting various handwriting parameters into multidimensional data. Recently, Carbune et al. [11] have developed an advanced handwriting recognition system at Google that leverages LSTM to support 102 languages. The researchers convert handwriting data to a 5-dimensional point including x-coordinate, y-coordinate, touchpoint timestamp, pen-up or pen-down, and stroke. Apart from this, they used Bézier Curves to present trajectories in space. Handwriting recognition finds useful applications in signature verification, document verification, etc.
A Novel Weighted SVM Classifier Based on SCA for Handwritten Marathi Character Recognition
Published in IETE Journal of Research, 2022
Surendra P. Ramteke, Ajay A. Gurjar, Dhiraj S. Deshmukh
The author [20] assesses the impacts of two sorts of character-level NNLMs as FNNLMs (feed forward neural network LMs) and RNNLMs (recurrent neural network LMs) for enhancing Chinese handwriting recognition. The hybrid LMs are built by joining both neural networks (FNNLMs and RNNLMs) with BLMs. For reasonable correlation with state-of-the-art system and BLMs, assessed in a system with the similar character over-segmentation and classification models as previously, different LMs are analyzed utilizing a little text corpus used previously. In their work the execution of both ICDAR-2013 and CASIA-HWDB competition dataset is enhanced essentially. The character level accurate rate and correct rate accomplish 95.88% and 95.95% in CASIA-HWDB, respectively. R Pramanik et al. [21] have introduced a novel shape decomposition-based segmentation technique to disintegrate the compound characters into prominent shape components. The less number of classes to recognize the shape decomposition lessens the classification complexity also enhances the recognition accuracy at the same time. At the segmentation area the decomposition is done where the two fundamental shapes are joined to form a compound character.
Recognition of handwritten characters from Devanagari, Bangla, and Odia languages using transfer-learning-based VGG-16 networks
Published in The Imaging Science Journal, 2022
Raghunath Dey, Rakesh Chandra Balabantaray, Sidharth Samanta
It is possible to employ handwriting recognition in academic and business applications. A wide range of handwriting styles and languages must be considered when trying to identify someone's handwriting. Indeed, handwriting recognition by deep learning algorithms has long been an influential application. In addition, many different handwriting styles exist that have a variety of ways to write a word. The characters are written in cursive in some cases, where they may be connected to each other. Handwritten characters are harder to interpret than printed characters when it comes to character recognition. Furthermore, the size and shape of handwritten characters made by different writers varied in several ways. The writing styles of different characters differ significantly, which makes identification difficult. Character identification becomes more difficult due to the presence of similarities in various different character classes, overlaps, and linkages between neighbouring characters. Handwritten characters are difficult to recognize due to the range of writing styles and authors, as well as the characters' complexity [1]. Convolutional neural network models are particularly effective in accurately identifying handwritten characters [2]. Furthermore, CNN does not require further feature extraction work.
Towards More Direct Text Editing With Handwriting Interfaces
Published in International Journal of Human–Computer Interaction, 2023
We used both text editors designed for the experiment. Further, we conducted a pilot study to determine a suitable font size for participants to manipulate text easily in both text editors; a mono-spaced 25 pt size, which is 7.9 mm on screen, was selected. An article occupied about half of the screen, therefore, we provided two articles at a time for the proofreading task, as shown in Figure 3. By closing the window and opening a new window, participants could access the next two articles. For the revision task, seven or eight samples were provided in a window. The Microsoft Ink library was used for online unconstrained handwriting recognition. The text editors were implemented with the C# Windows Presentation Foundation.