 |
3 books on Handwriting Recognition [PDF]
June 02, 2025 | 20 |
These books describe ML and older algorithms for handwriting recognition (neural networks, CTC, transformers), pre-processing of handwriting images (filtering, segmentation, feature extraction) and real-world cases: digitization of archives, recognition of questionnaires, checks, notes and medical records.
1. Optical Character Recognition Technologies and Algorithms
2025 by Richard Johnson
This book is mainly about OCR technologies, but also contains a short description of the latest achievements in handwriting recognition. And since handwriting recognition is not in great demand and is not very popular among book authors lately, we have to make do with this. In general, this book explains the unpopularity of the topic - due to the complexity of handwriting recognition and the small amount of data for training, which means poor output quality. But it also offers a solution - a hybrid system of CNN (convolutional neural network) trained on various printed fonts and RNN (recurrent neural network) for recognizing written characters with CTC (connected temporal classifier) - this is a special neural network training algorithm used when the input and output have different lengths and the exact alignment is unknown.
Download PDF
2. Handwriting Recognition: Fundamentals and Applications
2023 by Fouad Sabry
This book explores why handwriting recognition still sucks. In short - because its too difficult compared to pay-off. While even old OCR algorithms could digitalize printed text where the letters were physically separated, the intertwined characters of handwriting faced a problem known as Sayre's paradox. It goes like this: To correctly recognize a character/letter, a system needs to know where exactly its boundaries are. But to correctly determine the boundaries of a character, you need to already know what that character is. The first practical handwriting recognition software was written by Shelia Guberman in Moscow back in 1962. Then ParaGraph International and Lexicus developed handwriting recognition technologies in the early 1990s. Their algorithms were based on separating letters. The recognition engine was used to determine which digital symbol corresponded to each separated fragment. Feature extraction worked automatically, but developers still had to manually decide which features were more important. Because of this, the implementation took long time and mistakes because of recognizing two characters as one - constantly occurred. Newer methods of handwriting recognition are much more focused on identifying entire lines of text rather than on symbols. In particular, they rely on machine learning techniques that can learn visual characteristics without the need for manual feature engineering. To generate character probabilities, modern algorithms use convolutional networks that extract visual information from multiple overlapping windows of a text string image.
Download PDF
3. Fundamentals in Handwriting Recognition
2012 by Sebastiano Impedovo
This book describes the theoretical foundations of handwriting recognition systems. In particular, it explains how online recognition (when you write with a stylus on the screen of a smartphone) differs from offline recognition (when a program recognizes a handwritten paper). When the input is done online, it is easier for the program to build a flexible model, since it is possible to record dynamic patterns of the drawing. Thus, one of the main advantages of online devices is that the process itself provides a natural segmentation of the shape of characters into strokes. Offline recognition is more difficult not only because of the loss of dynamic information, but also because page scanning adds noise into the image. One of the main difficulties in recognizing words is the large variability in different handwriting samples obtained from different scribes. That is why in the field of offline recognition, the concept of regular and special features is used: a regular class includes features whose iconic representation is periodic; a special class consists of features that are accidents of regular features. Examples of regular features are textures and lines. Examples of special features are texture edges, endings, intersections, sharp angles and line forks. In this approach, after preprocessing, a regular part of the word is extracted. This part, also called the word axis is defined as the shortest path from one extreme (left) to the other (right), which remains inside the word body. Features are extracted by removing the axis from the word image.
Download PDF
How to download PDF:
1. Install Gooreader
2. Enter Book ID to the search box and press Enter
3. Click "Download Book" icon and select PDF*
* - note that for yellow books only preview pages are downloaded
1. Optical Character Recognition Technologies and Algorithms
2025 by Richard Johnson
This book is mainly about OCR technologies, but also contains a short description of the latest achievements in handwriting recognition. And since handwriting recognition is not in great demand and is not very popular among book authors lately, we have to make do with this. In general, this book explains the unpopularity of the topic - due to the complexity of handwriting recognition and the small amount of data for training, which means poor output quality. But it also offers a solution - a hybrid system of CNN (convolutional neural network) trained on various printed fonts and RNN (recurrent neural network) for recognizing written characters with CTC (connected temporal classifier) - this is a special neural network training algorithm used when the input and output have different lengths and the exact alignment is unknown.
Download PDF
2. Handwriting Recognition: Fundamentals and Applications
2023 by Fouad Sabry
This book explores why handwriting recognition still sucks. In short - because its too difficult compared to pay-off. While even old OCR algorithms could digitalize printed text where the letters were physically separated, the intertwined characters of handwriting faced a problem known as Sayre's paradox. It goes like this: To correctly recognize a character/letter, a system needs to know where exactly its boundaries are. But to correctly determine the boundaries of a character, you need to already know what that character is. The first practical handwriting recognition software was written by Shelia Guberman in Moscow back in 1962. Then ParaGraph International and Lexicus developed handwriting recognition technologies in the early 1990s. Their algorithms were based on separating letters. The recognition engine was used to determine which digital symbol corresponded to each separated fragment. Feature extraction worked automatically, but developers still had to manually decide which features were more important. Because of this, the implementation took long time and mistakes because of recognizing two characters as one - constantly occurred. Newer methods of handwriting recognition are much more focused on identifying entire lines of text rather than on symbols. In particular, they rely on machine learning techniques that can learn visual characteristics without the need for manual feature engineering. To generate character probabilities, modern algorithms use convolutional networks that extract visual information from multiple overlapping windows of a text string image.
Download PDF
3. Fundamentals in Handwriting Recognition
2012 by Sebastiano Impedovo
This book describes the theoretical foundations of handwriting recognition systems. In particular, it explains how online recognition (when you write with a stylus on the screen of a smartphone) differs from offline recognition (when a program recognizes a handwritten paper). When the input is done online, it is easier for the program to build a flexible model, since it is possible to record dynamic patterns of the drawing. Thus, one of the main advantages of online devices is that the process itself provides a natural segmentation of the shape of characters into strokes. Offline recognition is more difficult not only because of the loss of dynamic information, but also because page scanning adds noise into the image. One of the main difficulties in recognizing words is the large variability in different handwriting samples obtained from different scribes. That is why in the field of offline recognition, the concept of regular and special features is used: a regular class includes features whose iconic representation is periodic; a special class consists of features that are accidents of regular features. Examples of regular features are textures and lines. Examples of special features are texture edges, endings, intersections, sharp angles and line forks. In this approach, after preprocessing, a regular part of the word is extracted. This part, also called the word axis is defined as the shortest path from one extreme (left) to the other (right), which remains inside the word body. Features are extracted by removing the axis from the word image.
Download PDF
How to download PDF:
1. Install Gooreader
2. Enter Book ID to the search box and press Enter
3. Click "Download Book" icon and select PDF*
* - note that for yellow books only preview pages are downloaded
