Letter Segmentation Research Articles

A Study on Handwritten Text Recognition Classification using Diverse Deep Learning Techniques and Computation of CTC Loss

Handwriting recognition encompasses the conversion of hand-written text images into digital text, where input images yield predicted textual output. Optical Character Recognition (OCR) technology has conventionally fulfilled this role. With surging mobile phone usage, leveraging text detection via mobile cameras gains significance in fields like medical script processing and exam evaluation. To enhance image quality, noise reduction techniques like binarization and thresholding are applied. Image processing entails letter segmentation and extraction. In this paper, we propose a neural network classifier model amalgamating Convolution Neural Networks (CNNs) and Long Short-Term Memory (LSTM) networks. Leveraging an existing feature dataset, the model is trained. Input images are evaluated through the neural network, yielding recognized words in a text document format. Ultimately, the predicted output is derived via Connectionist Temporal Classification (CTC)- based loss computation. Keywords: Handwritten Recognition, Deep Learning Techniques, Optical Character Recognition.

Israel, an Anatomy of the World: Microcosmic Contours Between Romans 9–11 and 1–8

Abstract While a near consensus now deems Romans 9–11 integral to the letter rather than peripheral, explanations of how these chapters integrate remain disparate. Apparent links between Rom. 9–11 and 1–8 abound, but no one has yet exposed the underlying congruence between the two sections’ argumentative trajectories. Coordinating divine activity in these letter segments reveals the interpretive value of ‘microcosm’ imagery, an ancient and versatile metaphor that will prove more illuminating than ‘climax’ inter alia for understanding Paul’s presentation of Israel. In Romans, the κόσμος itself is an actor in the text, with Israel serving as its microcosm. Romans 9–11 then portrays God’s actions toward Israel in a microcosmic relation to God’s actions toward the κόσμος in Rom. 1–8. In its disobedience, judgement, and salvation, Israel embodies the anatomy of the world, a layered harmony that may help reconcile warring camps in recent Pauline scholarship.

Write, Attend and Spell

Text entry on a smartwatch is challenging due to its small form factor. Handwriting recognition using the built-in sensors of the watch (motion sensors, microphones, etc.) provides an efficient and natural solution to deal with this issue. However, prior works mainly focus on individual letter recognition rather than word recognition. Therefore, they need users to pause between adjacent letters for segmentation, which is counter-intuitive and significantly decreases the input speed. In this paper, we present 'Write, Attend and Spell' (WriteAS), a word-level text-entry system which enables free-style handwriting recognition using the motion signals of the smartwatch. First, we design a multimodal convolutional neural network (CNN) to abstract motion features across modalities. After that, a stacked dilated convolutional network with an encoder-decoder network is applied to get around letter segmentation and output words in an end-to-end way. More importantly, we leverage a multi-task sequence learning method to enable handwriting recognition in a streaming way. We construct the first sequence-to-sequence handwriting dataset using smartwatch. WriteAS can yield 9.3% character error rate (CER) on 250 words for new users and 3.8% CER for words unseen in the training set. In addition, WriteAS can handle various writing conditions very well. Given the promising performance, we envision that WriteAS can be a fast and accurate input tool for smartwatch.

RF-Wri: An Efficient Framework for RF-Based Device-Free Air-Writing Recognition

Machines are becoming indispensable in our lives and the requirements of the human-machine interactions are increasing. Conventional devices, such as a keyboard or touch screen, may not be preferred in future’s entertainment, home, and industrial applications. Device-free (non-contact) solutions will be even more popular. These solutions often use visual and acoustic technologies which have some disadvantages. The use of radio frequency (RF) waves for human-machine interaction such as air-writing (Wri), is a new and challenging problem. We propose a device-free machine learning-based air-writing recognition framework called <italic xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">RF-Wri</i> which can effectively distinguish 26 capital letters. Two-channel low-cost software-defined radios (SDR) and oppositely polarized antennas are used to provide polarization diversity which makes the accuracy of classification superior. Another critical novelty is the usage of Discrete Cosine Transform (DCT) coefficients as new features to represent RF waveform which provides writing speed and user invariant recognition. Discrete Wavelet Transform (DWT) filters and letter segmentation algorithm are used for signal de-noising and separating the air-writing activities, respectively. It is shown that Support Vector Machine (SVM) can successfully classify the measured RF waves of air-written letters. It is verified with various real measurements that the proposed framework, RF-Wri, achieves 95.15% accuracy in the classification of all 26 air-written letters and outperforms the fairly new WiFi-based air-writing recognition approaches.

Hijaiyah Letter Segmentation Using Connected Component Labeling Method

Arabic letters or Hijaiyah letters recognition is a challenge in itself because one letter consists of more than one character, namely the main character, companion character such as dots and lines, and punctuation called harakat. The image segmentation process is the most important in a character recognition system because it affects the separation of objects in an image. In this research, Hijaiyah letter segmentation aims to separate the letters according to the character of each letter using the Connected Component Labeling (CCL) method. Merging labels on each character will be done by looking for the Euclidean distance value from adjacent centroids. The experiment succeeded in segmenting each Hijaiyah character with an accuracy value of 86%.

Is there a cost at encoding words with joined letters during visual word recognition?

Abstract For simplicity, models of visual-word recognition have focused on printed words composed of separated letters, thus overlooking the processing of cursive words. Manso de Zuniga, Humphreys, and Evett (1991) claimed that there is an early “cursive normalization” encoding stage when processing written words with joined letters. To test this claim, we conducted a lexical decision experiment in which words were presented either with separated or joined letters. To examine if the cost of letter segmentation occurs early in processing, we also manipulated a factor (i.e., word-frequency) that is posited to affect subsequent lexical processing. Results showed faster response times for the words composed of separated letters than for the words composed of joined letters. This effect occurred similarly for low- and high-frequency words. Thus, the present data offer some empirical support to Manso de Zuniga et al.’s (1991) idea of an early “cursive normalization” stage when processing joined-letters words. This pattern of data can be used to constrain the mapping of the visual input into letter and word units in future versions of models of visual word recognition.

Trained convolutional neural network based on selected beta filters for Arabic letter recognition

AbstractThis paper presents a fast deep learning approach to segment and recognize off‐line Arabic printed and handwritten letters from words. We proposed a simple and powerful algorithm for Arabic letter segmentation based on vertical profile and baseline analysis. Then, we proposed a new method for feature extraction using fast wavelet transform. These extracted features are exploited as connection weights to build a convolutional neural network for each letter shape. Finally, all estimated model shapes are boosted to increase the robustness and performance of the proposed system. The proposed approach was tested on APTI and IESK‐arDB databases to evaluate performance for printed letters and handwritten letters, respectively. The obtained results show the robustness of our approach as well as the speed of the proposed recognition algorithm for both databases.This article is categorized under: Technologies &gt; Machine Learning Technologies &gt; Computational Intelligence Technologies &gt; Classification

Text Translation of Scanned Hindi Document to Braille via Image Processing

Objective: Braille - a model to reduce the illiteracy rate among the visually challenged people. So, this paper focuses on text translation/ conversion of scanned Hindi document to Braille, a single mode of communication for visually challenged people via Image processing techniques. Method/Analysis: The translation of scanned Hindi text into Braille code involves the following four processes: (i) Hindi database generation of consonants and matras using image segmentation. (ii) Segmentation of test images into lines followed by words and finally into letters. This step involves three types of segmentation: (a) line segmentation (b) word segmentation (c) letter segmentation. (iii) Letter matching of segmented letters with the generated Hindi database using Principal Component Analysis (PCA). (iv) Conversion of matched letter into corresponding Braille code. The algorithm takes test documents as input and reads the letter wise and maps the letters in the corresponding Braille code. Findings: Three Hindi documents containing single, multiple lines are taken as test documents. These test documents are successfully converted into their corresponding Braille code using the proposed algorithm. A lot of research work has been conducted which emphasizes the conversion of scanned Braille code into the text, but the conversion of scanned text to Braille has not been attempted so far. These results are a step further in the field of Braille communication, adding a new way through which Hindi literature can easily be made available to visually impaired people for improving their knowledge and easy access to documents and books as needed. Improvement/Application: The results obtained will provide easy access of Hindi literature to visually challenged people as there is limited availability of Braille textbooks in the market. The memory consumption is very less as the size of each image in Hindi database is 187x128 pixels. This work can be modified for real time applications, like instant conversion of text to Braille using handheld devices.

Texture Detection for Letter Carving Segmentation of Ancient Copper Inscriptions

As relics of history, ancient copper inscriptions are found in many countries. Information in the image or letter forms contained on copper ancient inscription has a very high value. The age and environmental factors caused damage to the surface of the inscription and also reduced the appearances of the image and letter. In this paper, we describe a novel segmentation methodology based on multi-texture features for ancient copper inscriptions which were severely damaged. The segmentation results of letters on ancient copper inscriptions by using the proposed method have an average accuracy of 90%. Based on these results, the proposed method is suitable for letter segmentation of the ancient copper inscriptions.

Arabic Character Recognition System Development

Abstract We develop Arabic Optical Character Recognition (AOCR) system that has five stages: preprocessing, segmentation, thinning, feature extraction, and classification. In preprocessing stage, we compare two skew estimation algorithms i.e. skew estimation by image moment and by skew triangle. We also implemented binarization and median filter. In thinning stage, we use Hilditch thinning algorithm incorporated by two templates, one to prevent superfluous tail and the other one to remove unnecessary interest point. In segmentation stage, line segmentation is done by horizontal projection cross verification by standard deviation, sub-word segmentation is done by connected pixel components, and letter segmentation is done by Zidouri algorithm. In the feature extraction stage, 24 features are extracted. The features can be grouped into three groups: main body features, perimeter- skeleton features, and secondary object features. In the classification stage, we use decision tree that generated by C4.5 algorithm. Functionality test showed that skew estimation using moment is more accurate than using skew triangle, median filter tends to erode the letter shape, and template addition into Hilditch algorithm gives a good result. Performance test yield these result. Line segmentation had 99.9% accuracy. Standard deviation is shown can reduce over-segmentation and quasi-line. Letter segmentation had 74% accuracy, tested on six different fonts. Classification components had 82% accuracy, tested by cross validation. Unfortunately, overall performance of the system only reached 48.3%.

Handwritten Text Image Authentication Using Back Propagation

Authentication is the act of confirming the truth of an attribute of a datum or entity. This might involve confirming the identity of a person, tracing the origins of an artefact, ensuring that a product is what it's packaging and labelling claims to be, or assuring that a computer program is a trusted one. The authentication of information can pose special problems (especially man-in-the-middle attacks), and is often wrapped up with authenticating identity. Literary can involve imitating the style of a famous author. If an original manuscript, typewritten text, or recording is available, then the medium itself (or its packaging - anything from a box to e-mail headers) can help prove or disprove the authenticity of the document. The use of digital images of handwritten historical documents has become more popular in recent years. Volunteers around the world now read thousands of these images as part of their indexing process. Handwritten text images of old documents are sometimes difficult to read or noisy due to the preservation of the document and quality of the image [1]. Handwritten text offers challenges that are rarely encountered in machine-printed text. In addition, most problems faced in reading machine- printed text (e.g., character recognition, word segmentation, letter segmentation, etc.) are more severe, in handwritten text. In this paper we Here in this paper we proposed a method for authenticating hand written text images using back propagation algorithm..

Cellular wave computer algorithms with spatial semantic embedding for handwritten text recognition

AbstractThe recognition of cursive handwritten texts is a complex, in some cases unsolvable, task. One problem is that in most cases it is difficult or impossible to identify each letter, even if the words are separated. In our new method, the identification of letters is not needed due to the extensive and iterative use of semantic and morphological information of a given language. We are using a spatial feature code, generated by a cellular nonlinear network (CNN) based cellular wave computer algorithm, and combine it with the linguistic properties of the given language. Most general‐purpose handwriting recognition systems lack the ability to integrate linguistic background knowledge because they use it only for post‐processing recognition results. The high‐level a priori background knowledge is, however, crucial in human reading and similarly it can boost recognition rates dramatically in case of recognition systems. In our new system we do not treat the visual source as the only input: geometric and linguistic information are given equal importance. On the geometric side we use word‐level holistic feature detection without letter segmentation by analogic CNN algorithms designed for cellular wave computers (IEEE Trans. Circuits Syst. 1993; 40:163–173; Cellular Neural Networks and Visual Computing, Foundations and Applications. Cambridge University Press: Cambridge, U.K., New York, 2002). The linguistic side is based on a morpho‐syntactic linguistic system (Proceedings of COLING‐2002, vol. II, Taipei, Taiwan, 2002; 1263–1267). A novel shape coding method is used to interface them, and their interaction is enhanced via an inverse filtering technique based on features that are global or of a low confidence value. A statistical context selection method is also applied to further reduce the output word lists. Copyright © 2008 John Wiley &amp; Sons, Ltd.

Hemispheric specialization for categorical and coordinate spatial relations during an image generation task: evidence from children and adults

Hemispheric specialization of categorical and coordinate image generation was assessed in adults, 8-year-old and 10-year-old children. In a standardized image generation task, participants decided whether probes, presented in a blank grid (categorical task) or bracketed square (coordinate task), would have appeared on a previously studied letter. To ensure that participants mentally generated the target letter, probe location was varied. “Early” probes appeared on letter segments that are first produced when the letter is drawn; while “late” probes appeared on later produced segments. Like previous adult studies, the grid task elicited a left hemisphere “categorical” strategy; while the bracket task elicits a right hemisphere “coordinate” strategy. However, contrary to previous research, the results reveal the significant and complex effects of probe location on categorical and coordinate image generation abilities. Specifically, early probes elicited a strong right hemisphere advantage for both tasks across all ages, whereas late probes produced a left hemisphere dissociation between categorical and coordinate processing. The left hemisphere dissociation was evident only for 10-year-olds and adults, suggesting that younger children are not yet proficient in generating spatial representations.

Alphabetical knowledge from whole words training: effects of explicit instruction and implicit experience on learning script segmentation

We investigated the possibility that pattern segmentation skills, specifically, phonological decoding, evolve implicitly in adult readers given training in an artificial script. In this Morse-like script each phoneme was represented by 2–3 discrete symbols. Subjects were trained in five consecutive sessions, on reading six nonsense words using a forced choice task that required translating symbol strings to sound patterns written in Latin letters. Three training conditions were compared within subject in terms of the time-course of learning and the ability to generalize the acquired knowledge (transfer): alphabetical whole words with letter decoding instruction (Explicit); alphabetical whole words (Implicit), and non-alphabetical whole words (Arbitrary). In separate blocks in each training session, a visual-matching task was administered using the same stimuli. Our results show: (a) that while all three training conditions were equally effective in terms of magnitude and time-course of learning accurate translation, each training condition resulted in a different type of knowledge (i.e. differential transfer). (b) Declarative knowledge of letters evolved from training on whole words only in subjects with previous experience in Explicit training. However, even with declarative knowledge of the specific letters subjects did not develop general letter segmentation skills. (c) Contrary to the robust transfer of learning gains to different stimuli within a given task, there was no significant transfer across tasks indicating that the locus of learning was task dependent. Altogether our results suggest that even given explicit letter instruction, training on word decoding may result in letter recognition rather than in alphabetic segmentation skills.

Network-based approach to online cursive script recognition

The idea of combining the network of HMMs and the dynamic programming-based search is highly relevant to online handwriting recognition. The word model of HMM network can be systematically constructed by concatenating letter and ligature HMM's while sharing common ones. Character recognition in such a network can be defined as the task of best aligning a given input sequence to the best path in the network. One distinguishing feature of the approach is that letter segmentation is obtained simultaneously with recognition but no extra computation is required.

Ligature modeling for online cursive script recognition

Online recognition of cursive words is a difficult task owing to variable shape and ambiguous letter boundaries. The approach proposed is based on hidden Markov modeling of letters and inter-letter patterns called ligatures occurring in cursive script. For each of the letters and the ligatures we create one HMM that models temporal and spatial variability of handwriting. By networking the two kinds of HMMs, we can design a network model for all words or composite characters. The network incorporates the knowledge sources of grammatical and structural constraints so that it can better capture the characteristics of handwriting. Given the network, the problem of recognition is formulated into that of finding the most likely path from the start node to the end node. A dynamic programming-based search for the optimal input-network alignment performs character recognition and letter segmentation simultaneously and efficiently. Experiments on Korean character showed correct recognition of up to 93.3% on unconstrained samples. It has also been compared with several other schemes of HMM-based recognition to characterize the proposed approach.

Implementierung von Diskreten und Semikontinuierlichen HMM in der Online-Handschrifterkennung

In this article a stochastic approach to online cursive script recognition is described. Hidden Markov Models (HMM) are trained on letter segments using discrete and semicontinous modeling for emission probability. Both variants are compared in the context of a letter based word recognizer. It is observed that in the context of the described preprocessing and feature extraction the recognition scores differ slightly with little advantage of the discrete type.

Is there really very rapid forgetting from primary memory? The role of expectancy and item importance in short-term recall.

In two experiments, subjects recalled one of two letter segments following a digit-filled retention interval. In Experiment 1, recall expectancy was manipulated by using precues that correctly informed or misinformed subjects concerning which letter segment would be tested for recall. In Experiment 2, item importance was varied by precuing one segment as important but requiring that the uncued segment be recalled first. Recall performance was very low under conditions of low expectancy and low segment importance, but the slopes of the retention functions did not demonstrate more rapid forgetting than under standard conditions. The previous observations of very rapid forgetting from primary memory may be a function of an elevated initial recall level in the earlier studies. Our retention functions were compared with predictions of the Estes perturbation model. The findings suggested that when secondary memory processes were reduced, forgetting order information from primary memory occurred at the same rate as that estimated on the basis of previous studies using the standard distractor task.

Off-line cursive script word recognition

Cursive script word recognition is the problem of transforming a word from the iconic form of cursive writing to its symbolic form. Several component processes of a recognition system for isolated offline cursive script words are described. A word image is transformed through a hierarchy of representation levels: points, contours, features, letters, and words. A unique feature representation is generated bottom-up from the image using statistical dependences between letters and features. Ratings for partially formed words are computed using a stack algorithm and a lexicon represented as a trie. Several novel techniques for low- and intermediate-level processing for cursive script are described, including heuristics for reference line finding, letter segmentation based on detecting local minima along the lower contour and areas with low vertical profiles, simultaneous encoding of contours and their topological relationships, extracting features, and finding shape-oriented events. Experiments demonstrating the performance of the system are also described. >

A multi-level perception approach to reading cursive script

Reading cursive script is the problem of transforming language from the iconic form of cursive human handwriting to one of ordinal representation. The problem involves elements of visual perception at one level of processing and those of language perception and understanding at a higher level. The problem is approached as one of multi-level perception in which a cursive script word image is transformed through a representational hierarchy. The levels are based on descriptions that use points, contours, features, letters, and words. Global control is hierarchical until an intermediate level after which it is heterarchical. A feature representation is generated bottom-up from the image using statistical dependencies between letters and features. Ratings for partially formed words are computed using a stack and a lexicon represented as a trie. Several heuristics for low- and intermediate-level processing for cursive script are introduced, including: reference-line finding using projection profile analysis, letter segmentation based on local lower contour minima and areas with low vertical profiles, simultaneous encoding of contours and their topological relationships, extracting features (e.g., middle loop, upper-zone stroke), and finding shape-oriented events. Two modes of learning are defined: initial training with user feedback and unsupervised adaptation to the writer. Experiments demonstrating the promise of the approach are described.