Implicit Segmentation Research Articles

An approach for handwritten Chinese text recognition unifying character segmentation and recognition

Text line recognition methods can be categorized into explicit segmentation based and implicit segmentation based ones. Explicit segmentation based methods require character-level annotation during training, while implicit segmentation based methods, trained on line-level annotated data, face alignment drift challenges. Though some methods have been proposed to address these challenges using weakly supervised object detection, they often rely on cumbersome pseudo-box generation processes and complex decoding. In this paper, we propose a unified framework to overcome these challenges, achieving high accuracy in text recognition and character segmentation. To eliminate the need of character-level annotated real text line data in training, we introduce a novel training paradigm that utilizes character-level annotated synthetic data and line-level annotated real data jointly. For synthetic data, candidate characters are explicitly aligned with labeled characters to generate hard labels for supervising model training. For real data, implicit alignment is produced by Connectionist Temporal Classification (CTC) mapping to provide soft labels for weakly-supervised model training. And for inference, we propose two decoding strategies leveraging the advantages of Non-Maximum Suppression (NMS) and CTC decoding. Extensive experiments on benchmark datasets demonstrate the superior performance of our method in text recognition and character localization, even with minimal amounts of character-level annotated line data.

Combination of explicit segmentation with Seq2Seq recognition for fine analysis of children handwriting

We consider the task of analyzing children handwriting in the context of a dictation task. The objective is to detect orthographic and phonological errors. To achieve this goal, we extend an existing handwriting analysis engine, based on an explicit segmentation of the handwritten input, originally developed for children copying exercises. We present a new approach, based on the combination of this analysis engine with a deep learning word recognition approach in order to improve both the recognition and segmentation performance. Explicit segmentation needs prior knowledge, and the deep network recognition predictions are a reliable approximation of the ground truth which can guide the analysis process. We propose to combine multiple prior knowledge strategies to further improve the analysis performance. Furthermore, we exploit the deep network approximate implicit segmentation to optimize the existing analysis process in terms of complexity.

Sub-sampling Approach for Unconstrained Arabic Scene Text Analysis by Implicit Segmentation based Deep Learning Classifier

The text extraction from the natural scene image is still a cumbersome task to perform. This paper presents a novel contribution and suggests the solution for cursive scene text analysis notably recognition of Arabic scene text appeared in the unconstrained environment. The hierarchical sub-sampling technique is adapted to investigate the potential through sub-sampling the window size of the given scene text sample. The deep learning architecture is presented by considering the complexity of the Arabic script. The conducted experiments present 96.81% accuracy at the character level. The comparison of the Arabic scene text with handwritten and printed data is outlined as well.

Multi-Classifier Jawi Handwritten Sub-Word Recognition

The problems and challenges in Jawi handwritten recognition are inherited from Arabic script which consists of cursive natures, large variety of writing styles due to its morphologically rich, ligature, overlapping characters, dialects and the low quality of the manuscripts images. The word segmentation is difficult because the existence of sub words due to the presence of space within words when contain disconnect characters. The performance of previous Jawi handwritten recognition still consider sub-par. There are three main problem of previous approach. First, the recognizer consist of multiple independent components where the improvement of performance in one component not shared across the systems. Secondly, the features extraction using features engineering approach only works on specific subsets of training data and is less capable to handle broader variants of testing data. Finally, the classifier used implicit segmentation where target class is sub-word with limited lexicon. This paper propose use of Deep Learning approach to address the first problem where training is conducted end-to-end from input to class output which enable the improvement of each component to improve overall performance. Secondly, Convolutional Network is use as learning features optimizes the data representation through end-to-end training of the parameters from raw input data to target class. Finally, A multi-classifier implicitly segments the sub-word into sequences of characters are proposed. The classifiers consists of one sub-word length classifier and seven character classifiers. This approach is lexicon-free to address absent of lexicon data. Experiments conducted on a Jawi handwritten standard dataset showed an accuracy of up to 92.20% and suggest that the approach used is superior to state-of-the-art methods of Jawi handwriting recognition.

A Hybrid Unsupervised Segmentation Algorithm for Arabic Speech Using Feature Fusion and a Genetic Algorithm (July 2018)

Speech is the most natural form of human communication. Major achievements have been made in developing systems that automatically recognize human speech and respond to it. An important preprocessing step in speech recognition systems (and other applications) is segmentation. Such a step is important in identifying the high-level semantics of speech sounds, including syllables, consonants, vowels…, and so on. There are basically two general approaches used for speech segmentation, namely, explicit and implicit (so-called with and without linguistic reference). Explicit segmentation uses a bottom-up process based on the concept of fixed-size frames. Such a framework is usually used on conjunction with hidden Markov models. The varying frame size or sample-by-sample approaches are mainly used in implicit segmentation techniques, which are based on the detection of spectral distortions. The main objective of this paper is to develop a novel speech segmentation algorithm for Arabic language. In this application, we focus on the accurate segmentation of Quran recitation. The proposed system starts with a set of initial segmentations using three basic speech features: entropy, zero crossings, and energy. The segmentation results obtained are then fused at the output level using a genetic algorithm-based optimization scheme. Together with the segmentation results, we also introduce the concept of speech units to model the fundamental entities representing Quran recitation. Our results show an enhanced performance in segmentation of about 20% over that of traditional single-feature-based segmentation techniques.

Urdu Nasta'liq text recognition using implicit segmentation based on multi-dimensional long short term memory neural networks.

The recognition of Arabic script and its derivatives such as Urdu, Persian, Pashto etc. is a difficult task due to complexity of this script. Particularly, Urdu text recognition is more difficult due to its Nasta’liq writing style. Nasta’liq writing style inherits complex calligraphic nature, which presents major issues to recognition of Urdu text owing to diagonality in writing, high cursiveness, context sensitivity and overlapping of characters. Therefore, the work done for recognition of Arabic script cannot be directly applied to Urdu recognition. We present Multi-dimensional Long Short Term Memory (MDLSTM) Recurrent Neural Networks with an output layer designed for sequence labeling for recognition of printed Urdu text-lines written in the Nasta’liq writing style. Experiments show that MDLSTM attained a recognition accuracy of 98% for the unconstrained Urdu Nasta’liq printed text, which significantly outperforms the state-of-the-art techniques.

Contextual factors in lexical processing: the case of French Liaison

ABSTRACTLower-level and higher-level processes during lexical recognition were investigated using ambiguous pseudo-noun cases related to liaisons in French. In phrases such as un onche and un nonche, the misalignment in the former liaison case produces an identical surface form as in the latter consonant-initial case, both [œ̃.nɔ̃ʃ], and two possible interpretations (onche, nonche) enter into competition. Quebec-French-speaking adults performed an implicit segmentation task testing the use of different factors. Results showed a dominant effect of syntactic category, with a general bias for vowel-initial interpretation when targets followed a determiner. The use of specific liaison acoustic cues for disambiguation was found for /z/ and /n/ only in adjective context. Liaison frequency and onset probability had no clear influence. Thus, the contextual knowledge of liaison-causing words is crucial for lexical recognition. These findings are consistent with the predictions of the hierarchy proposed by Mattys, White, and Melhorn (2005).

Offline cursive Urdu-Nastaliq script recognition using multidimensional recurrent neural networks

Optical Character Recognition of cursive scripts remains a challenging task due to a large number of character shapes, inter- and intra-word overlaps, context sensitivity and diagonality of text. This paper presents an implicit segmentation based recognition system for Urdu text lines in Nastaliq script. The proposed technique relies on sliding overlapped windows on lines of text and extracting a set of statistical features. The extracted features are fed to a multi-dimensional long short term memory recurrent neural network (MDLSTM RNN) with a connectionist temporal classification (CTC) output layer that labels the character sequences. Experimental study of the proposed technique is carried out on the standard Urdu Printed Text-line Images (UPTI) database which comprises 10,000 text lines in Nastaliq font. Evaluations under different experimental settings realize promising recognition rates with a highest character recognition rate of 96.40%.

Putty Capital's Shadow of the Invisible Hand on Clay Labor: On the Emergent Risk of Differing Speeds of European Union Capital and Labor Freedom

Globalization led to unprecedented risks stemming from global interconnectedness. Economic trade may distribute benefits of international exchange unevenly due to fundamental barriers of distance, national borders and implicit market segmentation. In order to equalize trade prosperity, the European Union (EU) 4 freedoms of goods, services, capital and labor are established by a neoliberal policy framework and the Eurozone featuring a common currency. While there is a vital central monetary union and since the 2008/09 World Financial Crisis a common European fiscal pact, EU free trade is limited regarding labor mobility. Explicit labor transfer constraints comprise of work permission requirements and sector specific restrictions while implicit drawbacks arise due to specific language, cultural and skill requirements. The paper addresses risks emerging from labor immobility in combination with more vital capital and goods movements. Within the European compound, full capital flows and export opportunities may gravitate trade benefits towards original EU core countries, while periphery countries that became later part of the EU are shunned from full employment productivity. A less mobile workforce in the EU periphery is described as a reserve army of labor with social problems invisible to the core union as for remaining out of focus due to national borders and geographic distance. Trade movements within the EU are analyzed with attention to export, unemployment as well as migration patterns in order to advocate for attention to full labor freedom within the EU following the greater goal of Ricardian fair and mutually-beneficial trade enabled through a harmonious interplay of national government and European governance polity.

Overlap and Separation: Religion-Work Boundaries in the Workplace

How do corporate professionals negotiate boundaries between work and religion in the workplace? While there has been considerable scholarship on the relationship between work-nonwork boundaries in the workplace and on religion and spirituality in the workplace, these two domains of inquiry remain largely disconnected. This paper attempts to bridge these conversations by developing new theory on the boundaries between religion and work in the workplace. It supports this framework by drawing on empirical data from 12 months of participant observation and 122 interviews in two nonwestern cities, Bangalore and Dubai. Building on Nippert-Eng’s (1996) phenomenology of boundary-work, I theorize three modes of overlap (fusion, instrumental overlap, and invasive overlap) and two modes of separation (implicit segmentation and explicit segmentation) of the experiential realms of religion and work in global corporate workplaces, and discuss the conditions that precipitate these configurations.

Statistical segmentation of streams of syllables: a pilot EEG study

Statistical segmentation of streams of syllables: a pilot EEG study

Implicit Segmentation of a Stream of Syllables Based on Transitional Probabilities: An MEG Study

Statistical segmentation of continuous speech, i.e., the ability to utilise transitional probabilities between syllables in order to detect word boundaries, is reflected in the brain's auditory event-related potentials (ERPs). The N1 and N400 ERP components are typically enhanced for word onsets compared to random syllables during active listening. We used magnetoencephalography (MEG) to record event-related fields (ERFs) simultaneously with ERPs to syllables in a continuous sequence consisting of ten repeating tri-syllabic pseudowords and unexpected syllables presented between these pseudowords. We found the responses to differ between the syllables within the pseudowords and between the expected and unexpected syllables, reflecting an implicit process extracting the statistical characteristics of the sequence and monitoring for unexpected syllables.

A case study of pricing strategies in European airline markets: The London – Amsterdam route

Published fares London–Amsterdam are used to examine the pricing practices of low-cost and legacy carriers when operating in a large and crowded market. We investigate two strategies of market segmentation involving the time before departure the ticket has been bought, inter-temporal segmentation, and the duration of the stay, implicit segmentation. We find inter-temporal price discrimination emerges as an important strategy for all pricing but the two legacy carriers involved, British Airways and KLM, differ in their use of stay restrictions; British Airways does not assign a specific role to the duration of stay, while KLM make use of such rules extensively in price setting.

Enhanced Volume Rendering Techniques for High-Resolution Color Cryo-Imaging Data.

We are developing enhanced volume rendering techniques for color image data. One target application is cryo-imaging, which provides whole-mouse, micron-scale, anatomical color, and molecular fluorescence image volumes by alternatively sectioning and imaging the frozen tissue block face. With the rich color images provided by cryo-imaging, we use true-color volume rendering and visually enhance anatomical regions by proper selection of voxel opacity. To compute opacity, we use color and/or gradient feature detection followed by suitable opacity transfer functions (OTF). An interactive user interface allows one to select from among multiple color and gradient feature detectors, OTF's, and their associated parameters, and to compute in live time new volume visualizations from within the Amira platform. We are also developing multi-resolution volume rendering techniques to accommodate extremely large (>60GB) cryo-image data sets. Together, these enhancements enable us to interactively interrogate cryo-image volume data and create useful renderings with "implicit segmentation" of organs.

Region growing based segmentation algorithm for typewritten and handwritten text recognition

This paper presents a new technique of high accuracy to recognize both typewritten and handwritten English and Arabic texts without thinning. After segmenting the text into lines (horizontal segmentation) and the lines into words, it separates the word into its letters. Separating a text line (row) into words and a word into letters is performed by using the region growing technique (implicit segmentation) on the basis of three essential lines in a text row. This saves time as there is no need to skeletonize or to physically isolate letters from the tested word whilst the input data involves only the basic information—the scanned text. The baseline is detected, the word contour is defined and the word is implicitly segmented into its letters according to a novel algorithm described in the paper. The extracted letter with its dots is used as one unit in the system of recognition. It is resized into a 9 × 9 matrix following bilinear interpolation after applying a lowpass filter to reduce aliasing. Then the elements are scaled to the interval [0,1]. The resulting array is considered as the input to the designed neural network. For typewritten texts, three types of Arabic letter fonts are used—Arial, Arabic Transparent and Simplified Arabic. The results showed an average recognition success rate of 93% for Arabic typewriting. This segmentation approach has also found its application in handwritten text where words are classified with a relatively high recognition rate for both Arabic and English languages. The experiments were performed in MATLAB and have shown promising results that can be a good base for further analysis and considerations of Arabic and other cursive language text recognition as well as English handwritten texts. For English handwritten classification, a success rate of about 80% in average was achieved while for Arabic handwritten text, the algorithm performance was successful in about 90%. The recent results have shown increasing success for both Arabic and English texts.

Virtual neuroendoscopy: MRI-based three-dimensional visualization of the cranial nerves in the posterior cranial fossa

This article presents the advances of three-dimensional (3D) virtual neuroendoscopy of the cranial nerves (CN) in the posterior fossa. Interactive 3D visualizations were generated and the anatomical landmarks, such as the root entry/exit zones (REZ) and cisternal segments of the CN were evaluated. Twenty patients (M:F, 6:14) with trigeminal neuralgia (TN) underwent MRI constructive interference in steady state (MRICISS) imaging and subsequent 3D visualization based on explicit segmentation of the MRICISS data and interactive evaluation with direct volume rendering including implicit segmentation. The 3D topography of the interesting CN V – X were evaluated with interactive and virtual neuroendoscopy. The anatomical landmarks of the CN V – X could be visualized in all 20 cases (100%). The systematic application of virtual neuroendoscopy could be realized in all patients for the non-invasive observation of the CN without any technical difficulties. Interactive 3D visualization using explicit and implicit techniques for segmentation, and 3D direct volume rendering is demonstrated to successfully identify 3D neurovascular relations in patients with trigeminal neuralgia. It has the ability to provide a useful tool for surgeons in the pre- and intraoperative evaluation of such cases.

MIXED DISCRIMINANT TRAINING OF HYBRID ANN/HMM SYSTEMS FOR ONLINE HANDWRITTEN WORD RECOGNITION

Online handwritten word recognition systems usually rely on Hidden Markov Models (HMMs), which are effective under many circumstances, but suffer some major limitations in real world applications. Artificial neural networks (ANN) appear to be a promising alternative, however they failed to model sequence data such as online handwriting due to their variable lengths. As a consequence, by combining HMMs and ANN, we can expect to take advantage of the robustness and flexibility of the HMMs generative models and of the discriminative power of the ANN. Training such a hybrid system is not straightforward, this is why so few attempts are encountered in literature. We compare several different training schemes: maximum likelihood (ML) and maximum mutual information (MMI) criteria in the framework of online handwriting recognition with a global optimization approach defined at the word level. A new generic criterion mixing generative model and discriminant trainings is proposed, it allows to train a multistate TDNN-HMM system directly at the word level. This architecture is based on an analytical approach with an implicit segmentation. To control the implicit segmentation and to initialize correctly the system without bootstrapping with another recognition system, we have defined a process that constraints the segmentation path and a measure called Average Segmentation Rate (ASR). Recognition experiments on the online IRONOFF database demonstrated the interest of the generic training criterion and the control of the implicit segmentation.

Analysis and 3-dimensional visualization of neurovascular compression syndromes 1

Rationale and Objectives. Neurovascular compression syndromes are currently examined with 2-dimensional representations of tomographic volumes. To overcome this drawback, coarse segmentation followed by direct volume rendering of magnetic resonance (MR) data is introduced supporting a detailed 3-dimensional analysis of the related structures. Materials and Methods. This approach is based on MR-CISS (constructive interference in steady state) volumes providing the required high resolution to achieve an improved spatial understanding. In relation to the size of the involved nerves and vessels, an explicit segmentation is extremely difficult. Therefore, a semi-automatic preprocessing sequence was developed consisting of noise reduction, morphologic filtering, and volume growing. To delineate the target structures within the segmented and labeled subvolumes, interactive direct volume rendering was applied that allows delineating the target structures in the area of the cerebrospinal fluid with implicit segmentation based on predefined transfer functions assigning opacity and color values to the intensity values of the image data. For a further improved analysis, registration of the MR-CISS volumes with MR angiography is recommended to support differentiating vessels and nerves on the one side and arteries and veins on the other. Results. The presented method was applied in a consecutive series of 47 cases of different neurovascular compression syndromes, supporting the presurgical analysis of the image data. Additionally, the results were compared with the operative findings. Conclusion. Overall, this approach contributes significantly to an optimized 3-dimensional analysis and understanding of neurovascular compression syndromes. Based on the obtained results, it is of high value for the planning of surgery.

The recognition of handwritten numeral strings using a two-stage HMM-based method

In this paper, a two-stage HMM-based recognition method allows us to compensate for the possible loss in terms of recognition performance caused by the necessary trade-off between segmentation and recognition in an implicit segmentation-based strategy. The first stage consists of an implicit segmentation process that takes into account some contextual information to provide multiple segmentation-recognition hypotheses for a given preprocessed string. These hypotheses are verified and re-ranked in a second stage by using an isolated digit classifier. This method enables the use of two sets of features and numeral models: one taking into account both the segmentation and recognition aspects in an implicit segmentation-based strategy, and the other considering just the recognition aspects of isolated digits. These two stages have been shown to be complementary, in the sense that the verification stage compensates for the loss in terms of recognition performance brought about by the necessary tradeoff between segmentation and recognition carried out in the first stage. The experiments on 12,802 handwritten numeral strings of different lengths have shown that the use of a two-stage recognition strategy is a promising idea. The verification stage brought about an average improvement of 9.9% on the string recognition rates. On touching digit pairs, the method achieved a recognition rate of 89.6%.

Sensitivity to Sonority for Print Processing in Normal Readers and Dyslexic Children

We investigated sonority effects on letters cohesion and possible implicit segmentation for printed CVCCVs in French. Adult readers, dyslexic children and control children had to recall the color of target letters in briefly presented two-coloured pseudowords. The distribution of colours was either compatible or not with the hypothesized underlying CVC.CV structure. Sonority either did or did not obey the phonotactic rule for optimum syllable sequence (higher sonority for the end of the first syllable than for the beginning of the subsequent one). Data showed that CVC.CV repartition was respected by adult readers (and, to a lesser extend, by dyslexic children) if this phonotactic rule was obeyed. Only dyslexic children associated the two consonants as an onset cluster (CV.CCV) if this rule was not obeyed. Therefore, dyslexic children exhibit sensitivity to sonority for implicit syllabification in reading, but the phonotactic rule about syllabic sequence is too strictly applied and CVC syllables with obstruent coda are refused.