No description available.

This thesis describes the implementation of a speaker dependent connected-digit recognizer using continuous Hidden Markov Modeling (HMM). The speech recognition system was implemented using MATLAB and on the ADSP-2181, a digital signal processor manufactured by Analog Devices. Linear predictive coding (LPC) analysis was first performed on a speech signal to model the characteristics of the vocal tract filter. A 7 state continuous HMM with 4 mixture density components was used to model each digit. The Viterbi reestimation method was primarily used in the training phase to obtain the parameters of the HMM. Viterbi decoding was used for the recognition phase. The system was first implemented as an isolated word recognizer. Recognition rates exceeding 99% were obtained on both the MATLAB and the ADSP-2181 implementations. For continuous word recognition, several algorithms were implemented and compared. Using MATLAB, recognition rates exceeding 90% were obtained. In addition, the algorithms were implemented on the ADSP-2181 yielding recognition rates comparable to the MATLAB implementation. / Master of Science

Objets de collection depuis les temps anciens, de nos jours les pièces de monnaie constituent un marché de plus en plus important. L’évaluation par des experts de l’état de conservation des pièces de monnaie, que nous nommons gradation, joue un rôle important pour déterminer leur valeur sur le marché. Dans le but de grader des pièces de monnaie de manière efficace et objective, la société GENI collabore avec le laboratoire LIRIS, afin d’automatiser le processus de gradation à partir de photos de pièces de monnaie.L’objectif principal de cette thèse est de fournir une aide à la gradation des pièces de monnaie à partir des photos de qualité. Le projet est composé de quatre phases : segmentation des monnaies, identification du type monétaire, détection et reconnaissance du millésime et gradation des monnaies.Dans la première phase, la pièce de monnaie est segmentée de sa photo de manière précise à l’aide d’un modèle paramétrique déformable. Ce dernier permet également d’extraire des caractéristiques de la pièce de monnaie telles que sa taille, son nombre de coins, de pans, etc.Lors de la deuxième phase, nous cherchons dans une base de données le type monétaire de référence correspondant à la pièce de monnaie requête à l’aide de scores de similarité basés sur des graphes. Le premier score se base sur des caractéristiques locales des contours en relief, et le second, qui est semi-global, permet de mettre en évidence des différences de motifs.La troisième phase concerne la reconnaissance du millésime. Il s’agit d’un sujet difficile car les caractères, dans ce contexte, ont un premier plan de couleur très similaire à l’arrière-plan. Après avoir localisé la zone du millésime et l’avoir découpée en imagettes de chiffres, nous proposons une méthode de reconnaissance de chiffres à l’aide de caractéristiques « topologiques ».Enfin, concernant la gradation des monnaies, nous proposons une méthode se basant sur une quantification des « éléments inattendus » comme les rayures et les taches. La pièce de monnaie est d’abord recalée sur une monnaie de référence, puis, nous détectons les « éléments inattendus » significatifs sur des zones d’intérêt. Enfin, concernant les « éléments inattendus » ténus difficiles à repérer individuellement, nous détectons les zones granuleuses à l’aide du Deep Learning. Le résultat obtenu par cette méthode, proche de ce que l’expert réalise « à la main », servira d’aide aux numismates. / Coins have been collected and studied since ancient times. Today, coin collection has become a hobby for anyone who wants to participate in. Coin grading is a way of determining the physical condition of a coin to provide an indicator for its market value. For grading coins on a large scale and in a relatively objective way, the GENI company cooperates with the laboratory LIRIS to automate the process by using coin photos.The main objective of this thesis is to grade coins from well-conditioned photos. The project is composed of four steps: coin segmentation from raw photos, monetary type identification, coin date detection and recognition and, coin grading.The first step is to extract coins from raw photos with a high precision. With a deformable geometric model, we segment precisely round coins, many-sided coins, wavy edged coins and holed coins by recognizing their shapes.The second step consists of coin recognition or monetary type identification. We match the query coin to the most identical type reference by using two similarity scores. The first similarity score is based on local features of relief contours. The second similarity score is a semi-global measure that highlights the difference between relief patterns.The third step is to detect and recognize coin dates. However, the fact that such characters have the same color as the background makes traditional optical character recognition methods difficult to apply. After extracting the date zone and cropping it into digit images, we propose a learning-free method to recognize those digits by analyzing their “topological” features.In the last step, the grading process is carried out by quantification of "unexpected elements" such as scratches and dirty marks. The coin to grade is registered to a reference coin. Then, large “unexpected elements” are detected in some regions of interest. However, some “micro-scratches” are difficult to extract individually but all together make a "grainy" surface. To deal with it, we use Deep Learning techniques to classify those grainy zones containing such “micro-scratches”. The result of our system, which is close to the manual expert one, is considered as a useful help for numismatists.

This thesis aims to implement a segmentation-free strategy in the context of handwritten multi-digit string recognition. Three models namely VGG-16, CRNN and 4C are built to be evaluated and benchmarked, also research about the effect of the different training set on model performance is carried out.

Support vector machines (SVMs) are a powerful tool for classification problems. SVMs have only been developed in the last 20 years with the availability of cheap and abundant computing power. SVMs are a non-statistical approach and make no assumptions about the distribution of the data. Here support vector machines are applied to a classic data set from the machine learning literature and the out-of-sample misclassification rates are compared to other classification methods. Finally, an algorithm for using support vector machines to address the difficulty in imputing missing categorical data is proposed and its performance is demonstrated under three different scenarios using data from the 1997 National Labor Survey.

In recent years neuromorphic computing systems have achieved a lot of success due to its ability to process data much faster and using much less power compared to traditional Von Neumann computing architectures. There are two main types of Artificial Neural Networks (ANNs), Feedforward Neural Network (FNN) and Recurrent Neural Network (RNN). In this thesis we first study the types of RNNs and then move on to Spiking Neural Networks (SNNs). SNNs are an improved version of ANNs that mimic biological neurons closely through the emission of spikes. This shows significant advantages in terms of power and energy when carrying out data intensive applications by allowing spatio-temporal information processing. On the other hand, emerging non-volatile memory (eNVM) technology is key to emulate neurons and synapses for in-memory computations for neuromorphic hardware. A particular eNVM technology, memristors, have received wide attention due to their scalability, compatibility with CMOS technology and low power consumption properties. In this work we develop a spiking neural network by incorporating an inter-spike interval encoding scheme to convert the incoming input signal to spikes and use a memristive crossbar to carry out in-memory computing operations. We develop a novel input and output processing engine for our network and demonstrate the spatio-temporal information processing capability. We demonstrate an accuracy of a 100% with our design through a small-scale hardware simulation for digit recognition and demonstrate an accuracy of 87% in software through MNIST simulations. / M.S. / In recent years neuromorphic computing systems have achieved a lot of success due to its ability to process data much faster and using much less power compared to traditional Von Neumann computing architectures. Artificial Neural Networks (ANNs) are models that mimic biological neurons where artificial neurons or neurodes are connected together via synapses, similar to the nervous system in the human body. here are two main types of Artificial Neural Networks (ANNs), Feedforward Neural Network (FNN) and Recurrent Neural Network (RNN). In this thesis we first study the types of RNNs and then move on to Spiking Neural Networks (SNNs). SNNs are an improved version of ANNs that mimic biological neurons closely through the emission of spikes. This shows significant advantages in terms of power and energy when carrying out data intensive applications by allowing spatio-temporal information processing capability. On the other hand, emerging non-volatile memory (eNVM) technology is key to emulate neurons and synapses for in-memory computations for neuromorphic hardware. A particular eNVM technology, memristors, have received wide attention due to their scalability, compatibility with CMOS technology and low power consumption properties. In this work we develop a spiking neural network by incorporating an inter-spike interval encoding scheme to convert the incoming input signal to spikes and use a memristive crossbar to carry out in-memory computing operations. We demonstrate the accuracy of our design through a small-scale hardware simulation for digit recognition and demonstrate an accuracy of 87% in software through MNIST simulations.

In this thesis we develop a system for recognition of strings of connected digits that can be used in a hands-free telephone system. We present a detailed description of the elements of the recognition system, such as an endpoint algorithm, the extraction of feature vectors from the speech samples, and the practical issues involved in training and recognition, in a Hidden Markov Model (HMM) based speech recognition system. We use continuous mixture densities to approximate the observation probability density functions (pdfs) in the HMM. While more complex in implementation, continuous (observation) HMMs provide superior performance to the discrete (observation) HMMs. Due to the nature of the application, ours is a speaker dependent recognition system and we have used a single speaker's speech to train and test our system. From the experimental evaluation of the effects of various model sizes on recognition performance, we observed that the use of HMMs with 7 states and 4 mixture density components yields average recognition rates better than 99% on the isolated digits. The level-building algorithm was used with the isolated digit models, which produced a recognition rate of better than 90% for 2-digit strings. For 3 and 4-digit strings, the performance was 83 and 64% respectively. These string recognition rates are much lower than expected for concatenation of single digits. This is most likely due to uncertainties in the location of the concatenated digits, which increases disproportionately with an increase in the number of digits in the string. / Master of Science

Throughout this project, we perform a multivariate analysis of the parameters of a long short-term memory (LSTM) system for handwritten digit recognition in order to understand the model’s behaviour. In particular, we are interested in explaining how this behaviour precipitate from its parameters, and what in the network is responsible for the model arriving at a certain decision. This problem is often referred to as the interpretability problem, and falls under scope of Explainable AI (XAI). The motivation is to make AI systems more transparent, so that we can establish trust between humans. For this purpose, we make use of the MNIST dataset, which has been successfully used in the past for tackling digit recognition problem. Moreover, the balance and the simplicity of the data makes it an appropriate dataset for carrying out this research. We start by investigating the linear output layer of the LSTM, which is directly associated with the models’ predictions. The analysis includes several experiments, where we apply various methods from linear algebra such as principal component analysis (PCA) and singular value decomposition (SVD), to interpret the parameters of the network. For example, we experiment with different setups of low-rank approximations of the weight output matrix, in order to see the importance of each singular vector for each class of the digits. We found out that cutting off the fifth left and right singular vectors the model practically losses its ability to predict eights. Finally, we present a framework for analysing the parameters of the hidden layer, along with our implementation of an LSTM based variational autoencoder that serves this purpose. / I det här projektet utför vi en multivariatanalys av parametrarna för ett long short-term memory system (LSTM) för igenkänning av handskrivna siffror för att förstå modellens beteende. Vi är särskilt intresserade av att förklara hur detta uppträdande kommer ur parametrarna, och vad i nätverket som ligger bakom den modell som kommer fram till ett visst beslut. Detta problem kallas ofta för interpretability problem och omfattas av förklarlig AI (XAI). Motiveringen är att göra AI-systemen öppnare, så att vi kan skapa förtroende mellan människor. I detta syfte använder vi MNIST-datamängden, som tidigare framgångsrikt har använts för att ta itu med problemet med igenkänning av siffror. Dessutom gör balansen och enkelheten i uppgifterna det till en lämplig uppsättning uppgifter för att utföra denna forskning. Vi börjar med att undersöka det linjära utdatalagret i LSTM, som är direkt kopplat till modellernas förutsägelser. Analysen omfattar flera experiment, där vi använder olika metoder från linjär algebra, som principalkomponentanalys (PCA) och singulärvärdesfaktorisering (SVD), för att tolka nätverkets parametrar. Vi experimenterar till exempel med olika uppsättningar av lågrangordnade approximationer av viktutmatrisen för att se vikten av varje enskild vektor för varje klass av siffrorna. Vi upptäckte att om man skär av den femte vänster och högervektorn förlorar modellen praktiskt taget sin förmåga att förutsäga siffran åtta. Slutligen lägger vi fram ett ramverk för analys av parametrarna för det dolda lagret, tillsammans med vårt genomförande av en LSTM-baserad variational autoencoder som tjänar detta syfte.

Context: The main purpose of this thesis is to build an automatic handwritten digit recognition method for the recognition of connected handwritten digit strings. To accomplish the recognition task, first, the digits were segmented into individual digits. Then, a digit recognition module is employed to classify each segmented digit completing the handwritten digit string recognition task. In this study, different machine learning methods, which are SVM, ANN and CNN architectures are used to achieve high performance on the digit string recognition problem. In these methods, images of digit strings are trained with the SVM, ANN and CNN model with HOG feature vectors and Deep learning methods structure by sliding a fixed size window through the images labeling each sub-image as a part of a digit or not. After the completion of the segmentation, to achieve the complete recognition of handwritten digits.Objective: The main purpose of this thesis is to find out the recognition performance of the methods. In order to analyze the performance of the methods, data is needed to be used for training using machine learning methods. Then digit data is tested on the desired machine learning technique. In this thesis, the following methods are performed: Implementation of HOG Feature extraction method with SVM Implementation of HOG Feature extraction method with ANN Implementation of Deep Learning methods with CNN Methods: This research will be carried out using two methods. The first research method is the ¨Literature Review¨ and the second ¨Experiment¨. Initially, a literature review is conducted to get a clear knowledge on the algorithms and techniques which will be used to answer the first research question i.e., to know which type of data is required for the machine learning methods and the data analysis is performed. Later on, with the knowledge of RQ1, Experimentation is conducted to answer the RQ2, RQ3, RQ4. Quantitative data is used to perform the experimentation because qualitative data which obtains from case-study and survey cannot be used for this experiment method as it contains non-numerical data. In this research, an experiment is conducted to find the best suitable machine learning method from the existing methods. As mentioned above in the objectives, an experiment is conducted using SVM, ANN, and CNN. By considering the results obtained from the experiment, a comparison is made on the metrics considered which results in CNN as the best method suitable for Documents Images. Results: Compare the results for SVM, ANN with HOG Feature extraction and the CNN method by using segmented results. Based on the Experiment results it is found that SVM and ANN have some drawbacks like low accuracy and low performance in the recognition of documented images. So, the other method i.e., CNN has greater performance with high accuracy. The following are the results of the recognition rates of each method. SVM performance - 39% ANN performance - 37% CNN performance - 71%. Conclusion: This research concentrates on providing an efficient method for recognition of automatic handwritten digits recognition. Here a sample training data is treated with existing machine learning and deep learning methods like SVM, ANN, and CNN. By the results obtained from the experimentation, it clearly is shown that the CNN method is much efficient with 71% performance when compared to ANN and SVM methods. Keywords: Handwritten Digit Recognition, Handwritten Digit Segmentation, Handwritten Digit Classification, Machine Learning Methods, Deep Learning, Image processing on document images, Support Vector Machine, Conventional Neural Networks, Artificial Neural Networks

No description available.