Figure 7 - uploaded by Rathin Radhakrishnan Nair
Content may be subject to copyright.
The figure showcases word error rate with each line representing the number of training samples from IAM dataset for the model and the Y-axis describes the number of fine-tune samples from Washington dataset
Similar publications
Neural networks have been widely used for high resource language (e.g. English) named entity recognition (NER) and have shown state-of-the-art results.However, for low resource languages, such as Dutch, Spanish, due to the limitation of resources and lack of annotated data, taggers tend to have lower performances.To narrow this gap, we propose thre...
In this paper, deep reinforcement learning (DRL) and knowledge transfer are used to achieve the effective control of the learning agent for the confrontation in the multi-agent systems. Firstly, a multi-agent Deep Deterministic Policy Gradient (DDPG) algorithm with parameter sharing is proposed to achieve confrontation decision-making of multi-agen...
Even though the assessment of the processes underlying Public Research Organizations and Industry (PRO-I) linkages suggested that interpersonal relationships are crucial for knowledge transfer and for the establishment of successful cooperative activities, only few studies have adopted a social network perspective to analyse the link between the ch...
Objectives: The main objectives of this paper are to determine the conditions of the transfer of knowledge in municipal
enterprises, to identify the subprocesses of knowledge transfer along with the methods used in their implementation,
and to determine the rules for knowledge transfer recommended for use in municipal enterprises.
Research Design &...
Citations
... In improving the performance of the model, the transfer learning method is commonly used as the initial weight initiation in the CNN algorithm. The use of transfer learning in the character dataset showcases the ability of the model to learn and adapt to a target dataset with limited training samples [6]. The use of transfer learning has also been carried out on handwritten text recognition and has shown good performance when trained in small databases [7]. ...
A text on an image often stores important information and directly carries high level semantics, makes it as important source of information and become a very active research topic. Many studies have shown that the use of CNN-based neural networks is quite effective and accurate for image classification which is the basis of text recognition. It can also be more enhanced by using transfer learning from pre-trained model trained on ImageNet dataset as an initial weight. In this research, the recognition is trained by using Chars74K dataset and the best model results then tested on some samples of IIIT-5K-Dataset. The research results showed that the best accuracy is the model that trained using VGG-16 architecture applied with image transformation of rotation 15{\deg}, image scale of 0.9, and the application of gaussian blur effect. The research model has an accuracy of 97.94% for validation data, 98.16% for test data, and 95.62% for the test data from IIIT-5K-Dataset. Based on these results, it can be concluded that pre-trained CNN can produce good accuracy for text recognition, and the model architecture that used in this study can be used as reference material in the development of text detection systems in the future
... The CNN model can be exploited in two different ways: either by designing the model manually or automatically and then achieving the training process from scratch [59][60][61][62] or by employing the Transfer Learning (TL) strategies to leverage features from off-the-shelf pre-trained models on bigger databases [30,63]. There are many available state-of-the-arts pre-trained CNN models that have been trained on the ImageNet database [64], such as VGGNet [65], GoogLeNet [66], ResNet [67], InceptionV3 [68], Xception [69], MobileNet [70], and DenseNet [71].The TL technique is used to transfer the acquired knowledge from one or more tasks in the source domain to another task in the target domain [28,63,[72][73][74][75] by utilizing a pre-trained network from a source domain that has a considerable amount of training data [74,76], and this helps in boosting the recognition accuracy or reducing training time [74]. The two widely utilized TL strategies are the feature extraction strategy from prior trained data and the fine-tuning strategy of the applied pre-trained network [63,77]. ...
Offline Arabic Handwriting Recognition (OAHR) has recently become instrumental in the areas of pattern recognition and image processing due to its application in several fields, such as office automation and document processing. However, OAHR continues to face several challenges, including the high variability of the Arabic script and its intrinsic characteristics such as cursiveness, ligatures, and diacritics, the unlimited variation in human handwriting, and the lack of large public databases. In this paper, we have introduced a novel context-aware model based on deep neural networks to address the challenges of recognizing offline handwritten Arabic text, including isolated digits, characters, and words. Specifically, we have proposed a supervised Convolutional Neural Network (CNN) model that contextually extracts optimal features and employs batch normalization and dropout regularization parameters to prevent overfitting and further enhance its generalization performance when compared to conventional deep learning models. We employed numerous deep stacked-convolutional layers to design the proposed Deep CNN (DCNN) architecture. The proposed model was extensively evaluated, and it was observed to achieve excellent classification accuracy when compared to the existing state-of-the-art OAHR approaches on a diverse set of six benchmark databases, including MADBase (Digits), CMATERDB (Digits), HACDB (Characters), SUST-ALT (Digits), SUST-ALT (Characters), and SUST-ALT (Names). Further comparative experiments were conducted on the respective databases using the pre-trained VGGNet-19 and Mobile-Net models; additionally, generalization capabilities experiments on another language database (i.e., MNIST English Digits) were conducted, which showed the superiority of the proposed DCNN model.
... This approach achieved low word error rate and character error rate for documents written in multiple languages, documents written by different authors as well as for a variety of written documents. A study by Nair et al. [14] presents handwriting text recognition using CNN and MDLSTM based approaches. ...
In the last decade, social media has become the most popular and ubiquitous facilitator for creating, and sharing content, accessing information, and connecting people around the world. Nowadays, social media content generation is rapidly growing in the Sri Lankan community. Some content generation was a threat to peace, which led to unpleasant situations in the country within the last two years. Content moderation is the best solution for this problem, but manually processing millions of posts is a difficult task. For automation, separation and identification of Sinhala and English character in social media images is an essential project in content identification. In this work, Sinhala and English characters are separately identified from image posts and video thumbnails which are on public Facebook pages. This paper introduces a novel algorithm to separate Sinhala and English characters. It has utilized topological features, contour-based technique, and water reservoir-based technique. The proposed algorithm achieved 93% accuracy. Separated characters are recognized using the Convolutional Neural Network in TensorFlow framework.
... Convolutional Neural Networks (CNN) have a high performance in computer vision and pattern recognition. Many approaches implemented them to tackle problems such as object detection [29], [30], identifying actions in images [31], and text recognition [32]. Ji et al. proposed a three-dimensional convolution on a CNN (3DCNN) architecture to analyze video data [33]. ...
Suspicious behavior is likely to threaten security, assets, life, or freedom. This behavior has no particular pattern, which complicates the tasks to detect it and define it. Even for human observers, it is complex to spot suspicious behavior in surveillance videos. Some proposals to tackle abnormal and suspicious behavior-related problems are available in the literature. However, they usually suffer from high false-positive rates due to different classes with high visual similarity. The PreCrime Behavior method removes information related to a crime commission to focus on suspicious behavior before the crime happens. The resulting samples from different types of crime have a high-visual similarity with normal-behavior samples. To address this problem, we implemented 3D Convolutional Neural Networks and trained them under different approaches. Also, we tested different values in the number-of-filter parameter to optimize computational resources. Finally, the comparison between the performance using different training approaches shows the best option to improve the suspicious behavior detection on surveillance videos.
... Hence, an ANN can be configured for an application, such as data classification or character recognition through the learning process. The learning process involves adjusting the system to a connection [24]. The artificial neural network comprises a network of multiple simple processors, each with a small amount of local memory [39]. ...
Due to increased usage of digital technologies in all sectors and in almost all day to day activities to store and pass information, Handwriting character recognition has become a popular subject of research. Handwriting remains relevant, but people still want to have Handwriting copies converted into electronic copies that can be communicated and stored electronically. Handwriting character recognition refers to the computer's ability to detect and interpret intelligible Handwriting input from Handwriting sources such as touch screens, photographs, paper documents, and other sources. Handwriting characters remain complex since different individuals have different handwriting styles. This paper aims to report the development of a Handwriting character recognition system that will be used to read students and lectures Handwriting notes. The development is based on an artificial neural network, which is a field of study in artificial intelligence. Different techniques and methods are used to develop a Handwriting character recognition system. However, few of them focus on neural networks. The use of neural networks for recognizing Handwriting characters is more efficient and robust compared with other computing techniques. The paper also outlines the methodology, design, and architecture of the Handwriting character recognition system and testing and results of the system development. The aim is to demonstrate the effectiveness of neural networks for Handwriting character recognition.
... Convolutional Neural Networks (CNN) have a high performance in computer vision and pattern recognition. Many approaches implemented them to tackle problems such as object detection [29,30], identifying actions in images [31], and text recognition [32]. Ji et al. [33] proposed a three-dimensional convolution on a CNN (3DCNN) architecture to analyze video data. ...
Suspicious behavior is likely to threaten security, assets, life, or freedom. This behavior has no particular pattern, which complicates the tasks to detect it and define it. Even for human observers, it is complex to spot suspicious behavior in surveillance videos. Some proposals to tackle abnormal and suspicious behavior-related problems are available in the literature. However, they usually suffer from high false-positive rates due to different classes with high visual similarity. The Pre-Crime Behavior method removes information related to a crime commission to focus on suspicious behavior before the crime happens. The resulting samples from different types of crime have a high-visual similarity with normal-behavior samples. To address this problem, we implemented 3D Convolutional Neural Networks and trained them under different approaches. Also, we tested different values in the number-of-filter parameter to optimize computational resources. Finally, the comparison between the performance using different training approaches shows the best option to improve the suspicious behavior detection on surveillance videos.
... Such data hungry approaches have been commonly trained with the largest publicly available datasets, and then fine-tuned to the target collection to be recognized. Such tuning strategies [2,11,23] guarantee that the neural networks can be properly trained, ending up extracting relevant features from handwriting strokes, that are later revamped to the target collection. But fine-tuning presents the downside of needing manual annotations both from the source and target datasets. ...
... GW, with a disparate style. We compare against the recent works proposed by Nair et al. [23] and Arandillas et al. [2]. They achieve CER values of 59.3% and 82%, respectively with their models trained on IAM and tested over the GW test set. ...
... Such data hungry approaches have been commonly trained with the largest publicly available datasets, and then fine-tuned to the target collection to be recognized. Such tuning strategies [2,11,23] guarantee that the neural networks can be properly trained, ending up extracting relevant features from handwriting strokes, that are later revamped to the target collection. But fine-tuning presents the downside of needing manual annotations both from the source and target datasets. ...
... GW, with a disparate style. We compare against the recent works proposed by Nair et al. [23] and Arandillas et al. [2]. They achieve CER values of 59.3% and 82%, respectively with their models trained on IAM and tested over the GW test set. ...
Handwritten Text Recognition (HTR) is still a challenging problem because it must deal with two important difficulties: the variability among writing styles, and the scarcity of labelled data. To alleviate such problems, synthetic data generation and data augmentation are typically used to train HTR systems. However, training with such data produces encouraging but still inaccurate transcriptions in real words. In this paper, we propose an unsupervised writer adaptation approach that is able to automatically adjust a generic handwritten word recognizer, fully trained with synthetic fonts, towards a new incoming writer. We have experimentally validated our proposal using five different datasets, covering several challenges (i) the document source: modern and historic samples, which may involve paper degradation problems; (ii) different handwriting styles: single and multiple writer collections; and (iii) language, which involves different character combinations. Across these challenging collections, we show that our system is able to maintain its performance, thus, it provides a practical and generic approach to deal with new document collections without requiring any expensive and tedious manual annotation step.
... For example, in [2], this method focuses on writer adaptive handwriting recognition system i.e. it depends on the styles of writing of a particular author and hence focuses to construct a model that increases the efficiency of the generic recognition model for a definite author. In [4], a technique for pre-processing is proposed along with normalizing the data and also an SVM classifier based optical character recognition (OCR) .It precisely focuses on recognizing different styles of writing .Similarly, another paper [5] depicts an end-to-end text recognition technique that is capable of detecting text arranged in different lines in images. ...
