Dissertations / Theses on the topic 'Handwritten Bangla digit recognition'

We present an original algorithm for recognizing handwritten digits. We begin by introducing a virtually infinite collection of binary geometric features. The features are queries that ask if a particular geometric arrangement of local topographic codes is present in an image. The codes, which we call "tags", are too coarse and common to be informative by themselves, but the presence of geometric arrangements of tags ("tag arrangements") can provide substantial information about the shape of an image. Tag arrangements are features that are well-suited for handwritten digit recognition as their presence in an image is unaffected by a large number of transformations that do not affect the class of the image. It is impossible to calculate all of the features in an image. We therefore use decision trees to simultaneously determine a small collection of informative features and construct a classifier. By only considering a small random sample of queries at each mode we are able to generate multiple, randomized trees that determine a more varied and informative collection of features than is possible with a single tree. The trees, which provide posterior estimates of the class probabilities, are aggregated to produce a stable and robust classifier. We analyze the performance of this method and propose several means of augmenting its performance. Most notably, we introduce a nearest neighbor final test that reduces the already low error rate an additional 20-30%. Testing was done on a subset of a National Institute of Standards and Technology database, and we report a classification rate of 99.6%, comparable to the top results reported elsewhere.

碩士<br>淡江大學<br>電機工程學系<br>88<br>Optical character recognition (OCR) research dates back to 1950''s. In the late 1980''s, increasing computer-power generated renewed interest in OCR for the unrestricted machine-printed characters and handwritten characters. These applications widely range from automatic mail processing, automatic data entry into large administrative systems, license-plate identification, banking, automatic cartography, to reading devices for blind. These OCR methods generally fall into three categories: statistical methods, syntactic methods, and neural-fuzzy methods. Basically, flexibility and speed are the main features that characterize a good OCR system. In addition, selection of a feature extraction method is probably the most important factor in achieving high recognition performance for OCR systems. In this thesis, two different recognition methods were proposed. We first adopt a simple pattern recognition system based on HyperRectangular Composite Neural Networks (HRCNNs) incorporated with fuzzy logic to solve the handwritten digit recognition problem. The proposed recognition system tried to use features as fewer as possible, while to achieve correct recognition rate as higher as possible. The second method is based on the use of deformable models. The basic idea in using deformable models for digit recognition is that each digit has a model, and a test image is classified by finding the model which is most likely to have generated it. The quality of the match between model and test image depends on the deformation of the model. Here we show that by using neural networks to provide efficient deformation, the matching time can be significantly reduced. In addition, a new matching measure between the two images was also proposed to improve the recognition performance. These two methods are demonstrated on a handwritten digit recognition task.

碩士<br>國立臺灣大學<br>電機工程學研究所<br>102<br>Human-computer interaction (HCI) has been a popular research field recently. Hand gesture recognition is an important part of HCI that provides a natural way of communication. Handwritten recognition is a part of hand gesture recognition that provides an alternative method to input characters. In this thesis, we propose a handwritten recognition system to input English characters and digits without using traditional input devices such as keyboards and mice. Accuracy and real time processing are highly desired in the handwritten digit and character recognition of HCI. In order to improve the accuracy, we suggest a new feature extracting algorithm which contains the temporal and spatial information of hand writing paths. Furthermore, we use support vector machines and random forests to carry out feature classification. Experimental results show that the proposed method has a very high accuracy in the handwritten digit and character recognition in real time.

碩士<br>佛光人文社會學院<br>資訊學系碩士班<br>94<br>We employ Recurrent Neural Networks (RNNs) to create the short term memory between digits of postal code by arranging in groups handwritten digits and image processing techniques to reach the goal of recognizing them in real time. Furthermore, we can check the correctness of input digits and predict the next one that could show up by standing on the memory. By providing the prediction, users can choose one of these digits and detect the mistake. The system can train the postal code on-line in order to create the usual combination of the three digits and forecast them, if they are uncreated. In first phase of our study, by using the Multilayer Feedforward Artificial Neural Networks (MFANNs), the recognition rate for all the postal code in Taiwan is 92.12% before building the correlations of inter-digits. Second, in the building correlations of inter-digits phase, we can detect the possible mistake of handwritten digits in the midst of process by training the RNNs with all the postal code. At last, we treat the forward part of all the postal code as the base data in the on-line training phase. According to the experiment results, the system can add the postal code into memory randomly and combine them to form the usual wordbook.

碩士<br>國立中正大學<br>資訊工程研究所<br>107<br>The deep belief network (DBN) is implemented in this thesis. First, the MNIST database is used as a functional verification of the network architecture. Later, the relevant database applications for voice identification will be applied. In the training model extraction and hardware verification, the Matlab simulation is performed to determine the appropriate network size and layers which can achieve satisfactory identification results. Subsequently, the trained model is stored in ROMs and integrated into the proposed hardware architecture. Then the test data of the MNIST database are used to verify the accuracy of the DBN hardware circuit. With the development of artificial intelligence, researches on speech recognition and deep learning become increasingly popular. With the aging society, the hearing aids also attracting attention. Traditional hearing aids are susceptible to environmental sounds. In addition, the hearing aids need to be wore for a long time, and the design of the assistive devices needs to be light and low-power consumption. Therefore, deep learning is used for the environmental sound field in the hearing aids to improve the resistance to the environmental sound field, and suitable hearing compensation can be applied. Using application specific IC (ASIC) to implement audio equipment for hearing aids can achieve lightweight and low-power consumption, and is a trend in the design of hearing aids. To future, the relevant database applications for voice identification will be applied. Furthermore, reducing the access times of the external memory, dynamically adjust the accuracy of calculations and reducing the number of MACs to reduce the power consumption of the overall operation will be gradually studied in the future works.

碩士<br>國立交通大學<br>電子研究所<br>107<br>“Artificial Intelligence” becomes a hot topic in recent years. The word of “Artificial Intelligence”, AI, was coined by John McCarthy at the Dartmouth Conference in 1956. Due to the financial setbacks and the bottleneck of the algorithm, there were two periods of declining funding and interest in AI, also called as “AI winter”. In 2012, AlexNet, an eight-layer convolutional neural network (CNN) algorithm won the champion in the ImageNet Large Scale Visual Recognition Challenge. In 2016, AlphaGo, a program developed by Google DeepMind, beat the world champion of the Go, Lee Sedol. These successes laid important fundaments for AI nowadays. Neuromorphic computing is a promising approach to implement AI, which imitates the computation in human brain for achieving low-power consumption, parallel computing and fault tolerance. The neural network comprise a large amount of synapses and neurons, and the signal transmits from pre-neurons to synapses and then to post-neurons. The connection strength between synapses is not fixed because of the synaptic plasticity. The weight is adjusted by the Spike-Timing Dependent Plasticity (STDP) to allow learning in the network. Coincidentally, the analog resistive random-access memory (RRAM) shares the similar characteristics with the synapse. That is, the conductance change of RRAM can be modulated by different input voltages. Additionally, thanks to the simple structure of RRAM, it can be used to realize high density array and accelerate the neuromorphic computing. In this thesis, we implement the two-layer back propagation algorithm to recognize the enlarged patterns by using analog RRAM. According to the algorithm, we design the hardware to test the enlarged patterns. Besides, we discuss the device characteristic and its impact on the neural network. Finally, we demonstrate the recognition results of the enlarged patterns. There are five chapters in this thesis, and the main content of the research is described from Chapter 2 to Chapter 4. In chapter 2, we introduce the system architecture and the corresponding application of each sub-module. In a circuit system corresponding to a neural network, the RRAM crossbar array acts as the synaptic network, the pulse generator acts as the pre-neuron, the leaky integrate-and-fire circuit acts as the post-neuron, and the FPGA is the controller to control the clock or adjust the weight in the network. Then, we will introduce the sequence of each operation, the relationship between the RRAM and the peripheral circuits, and the precise waveforms in each operation. In chapter3, we analyze the characteristics of the analog RRAM. First, we test the characteristics of the single cell in DC and AC mode, the operation condition, the endurance and the line resistance. Second, because numerous cells of RRAM with acceptable characteristics are needed to implement the neural network, we design a rapid device screening method before and after device package. Then, we test the impact of breakdown devices in the array and discuss how to use the array without a perfect yield. Finally, we discuss the impacts on the RRAM when exposed to high-energy X-ray. In chapter 4, we implement the training and testing of enlarged patterns into the hardware, and modify the non-ideal peripheral circuits to improve the resolution of the system. Then, we demonstrate the results of pattern recognition from the two-layer neural network (100ｘ10ｘ 3). In this research, we implement the two-layer back propagation algorithm to recognize the enlarged patterns in the hardware. This makes the implement of the neuromorphic computing hardware closer to real applications. We believe the neuromorphic computing is capable of learning more complex tasks and making accurate decisions in the future.

碩士<br>國立臺灣師範大學<br>電機工程研究所<br>102<br>Human-computer interaction has been much popular in recent years, especially in the field of entertainment and education. In our previous work, we presented a method of Kinect-based mid-air handwritten digit recognition for a potential application to TV remote controllers. Nevertheless, its recognition accuracy is only about 86.7%. In this thesis, we propose an improved method based on multiple segments and scaled coding. Experimental results show that the proposed method can elevate the accuracy up to 94.6%. Furthermore, we also present two applications combined with our improved method on Android smart phones. This system can be applied to dial without touching the smart phone. This can be implemented in the kitchen or hospital when user's hands are unable to touch the screen.