Accuracy Of Recognition System Research Articles

An efficient character recognition method using enhanced HOG for spam image detection

Generally, a spam image is an unsolicited message electronically sent to a wide group of arbitrary addresses. Due to attractiveness and more difficult detection, spam images are the most complicated type of spam. One of the ways to encounter the spam images is an optical character recognition, OCR, method. In this paper, the proposed enhanced HOG feature extraction method has been used so that the optical character recognition system of spam has been enhanced by using the HOG feature extraction method in such a way to be both resistant against the character variations on scale and translation and to be computationally cost-effective. For these purposes, two steps of the cropped image and input image size normalization have been added to pre-processing stages. Support vector machine, SVM, was employed for classification. Two heuristic modifications including thickening of the thin characters in the pre-processing stage and non-discrimination in detecting the uppercase and lowercase letters with the same shapes in the classification stage have been also proposed to increase the system recognition accuracy. In the first heuristic modification, when all pixels of the output image are empty (the character is eliminated), the original image was made thicker by one layer. In the second modification, when recognizing the letters, no differentiation was considered between the uppercase and lowercase letters with the same shapes. An average recognition accuracy of the modified HOG method with two heuristic modifications equals 91.61% on Char74K database. Then, an optimum threshold for classification was investigated by ROC curve. The optimal cutoff point was 0.736 with the highest average accuracy, 94.20%, and AUC, area under curve, for ROC and precision–recall, PR, curves were 0.96 and 0.73, respectively. The proposed method was also examined on ICDAR2003 database, and the average accuracy and its optimum using ROC curve were 82.73% and 86.01%, respectively. These results of recognition accuracy and AUC for ROC and PR curve showed an outstanding enhancement in comparison with the best recognition rate of the previous methods.

EERA-KWS: A 163 TOPS/W Always-on Keyword Spotting Accelerator in 28nm CMOS Using Binary Weight Network and Precision Self-Adaptive Approximate Computing

This paper proposed an energy-efficient reconfigurable accelerator for keyword spotting (EERA-KWS) based on binary weight network (BWN) and fabricated in 28-nm CMOS technology. This keyword spotting system consists of two parts: the feature extraction based on melscale frequency cepstral coefficients (MFCC) and the keywords classification based on a BWN model, which is trained through the Google’s Speech Commands database and deployed on our custom. To reduce the power consumption while maintaining the system recognition accuracy, we first optimize the MFCC implementation with approximate computing techniques, including Pre-emphasis coefficient transformation, rectangular Mel filtering, Framing and FFT optimization. Then, we propose a precision self-adaptive reconfigurable accelerator with digital-analog mixed approximate computing units to process the BWN efficiently. Based on the SNR prediction of background noise and post-detection of network output confidence, the BWN accelerator data path can be dynamically and adaptively reconfigured as 4, 8, or 16 bits. For the BWN accelerator, we proposed a time-delay based addition unit to process bit-wise approximate computing for the convolution layers and fully connected layers, and a LUT based unit for the activation layers. Implemented under TSMC 28 nm HPC+ process technology, the estimated power is <inline-formula xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink"> <tex-math notation="LaTeX">$77.8~\mu \text{W}~\sim ~115.9\mu \text{W}$ </tex-math></inline-formula> , the energy efficiency can achieve 163 TOPS/W, which is over <inline-formula xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink"> <tex-math notation="LaTeX">$1.8\times $ </tex-math></inline-formula> better than the state-of-the-art architecture.

An Ultra-Low Power Always-On Keyword Spotting Accelerator Using Quantized Convolutional Neural Network and Voltage-Domain Analog Switching Network-Based Approximate Computing

An ultra-low power always-on keyword spotting (KWS) accelerator is implemented in 22nm CMOS technology, which is based on an optimized convolutional neural network (CNN). To reduce the power consumption while maintaining the system recognition accuracy, we first perform a bit-width quantization method on the proposed CNN to reduce the data/weight bit width required by the hardware computing unit without reducing the recognition accuracy. Then, we propose an approximate computing architecture for the quantized CNN using voltage-domain analog switching network based multiplication and addition unit. Implementation results show that this accelerator can support 10 keywords real time recognition under different noise types and SNRs, while the power consumption can be significantly reduced to 52 μW.

Asynchronous and Event-Based Fusion Systems for Affect Recognition on Naturalistic Data in Comparison to Conventional Approaches

Throughout many present studies dealing with multi-modal fusion, decisions are synchronously forced for fixed time segments across all modalities. Varying success is reported, sometimes performance is worse than unimodal classification. Our goal is the synergistic exploitation of multimodality whilst implementing a real-time system for affect recognition in a naturalistic setting. Therefore we present a categorization of possible fusion strategies for affect recognition on continuous time frames of complete recording sessions and we evaluate multiple implementations from resulting categories. These involve conventional fusion strategies as well as novel approaches that incorporate the asynchronous nature of observed modalities. Some of the latter algorithms consider temporal alignments between modalities and observed frames by applying asynchronous neural networks that use memory blocks to model temporal dependencies. Others use an indirect approach that introduces events as an intermediate layer to accumulate evidence for the target class through all modalities. Recognition results gained on a naturalistic conversational corpus show a drop in recognition accuracy when moving from unimodal classification to synchronous multimodal fusion. However, with our proposed asynchronous and event-based fusion techniques we are able to raise the recognition system's accuracy by 7.83 percent compared to video analysis and 13.71 percent in comparison to common fusion strategies.

A spatiotemporal approach for vision-based hand gesture recognition using Hough transform and neural network

Recognition of hand gesture is a crucial task as far as accuracy of recognition and real-time performance is concerned. The recognition accuracy of hand gesture systems used in applications from virtual reality to sign language recognition depends not only on the features extracted but also on the soft computing techniques used to recognize the gesture. This requires recognition system, with robust features which are unaffected by rotation, scaling, translation, variation in illumination and view of the gesture within the image. Accordingly, the research work presented in this paper aims to recognize the dynamic hand gestures using Hough transform-based spatiotemporal feature extraction for preprocessing and artificial neural network for identification. The system is implemented and tested using benchmark Cambridge hand gesture database and Sebastien dynamic hand posture database. As many as 32 features comprising of 19 Fourier descriptors, 3 geometrical features and 10 temporal features are obtained on each gesture sequence and are given as input to NN. The experimental results revealed that this approach is able to recognize dynamic hand gestures with an average recognition rate (RR) higher than 94% on Cambridge database which is better than that reported by earlier researchers while 98% RR on Sebastien database. The noteworthy feature of the proposed approach is that it relieves from the burden of foreground detection, generally required for other video processing systems.

Feature-Based Hand Gesture Recognition Using an FMCW Radar and its Temporal Feature Analysis

In this paper, feature-based gesture recognition in a frequency modulated continuous wave (FMCW) radar system is introduced. We obtain a range-Doppler map (RDM) from raw signals of FMCW radar and generate a variety of features from the RDM. The features are broadly defined to reflect radar-specific characteristics as well as statistical values commonly used in machine learning. Among these radar features, those that are highly correlated with gesture recognition are selected by the proposed feature selection algorithm, which is a wrapper-based feature selection algorithm incorporated with a quantum-inspired evolutionary algorithm (QEA). Furthermore, the information factor based on the minimum redundancy maximum relevance criterion is applied to QEA in order to find feature subsets effectively. The proposed algorithm is able to extract from all feature sets feature subsets related to gesture recognition, and improves the gesture recognition accuracy of the FMCW radar system. In addition, we analyze which features of the radar are helpful for gesture recognition and perform effective gesture recognition using the features determined through feature analysis.

Metric-learning-based high-discriminative local features extraction for iris recognition

Biometric-based authentication system is one of the main strategies to protect and control the access of users to important resources in any system and organization. Iris pattern is one of the best and most reliable biological features used in these systems. Extraction of high-discriminative local features can increase the recognition accuracy of iris-based biometric systems, especially when the number of users is high. Most of the existing methods utilize a combination of simple handcraft local feature models that deteriorate system performance when the number of users is increased. In this paper, after identification and segmentation of iris region, a new learning-based method is proposed to define and extract rotation- and illumination-invariant main local patterns associated with the iris texture. Afterwards, the metric-learning-based transform is employed to improve the discrimination of these patterns in recognition process. The proposed method was applied on more than 10,000 images from CASIA-V4, UBIRIS and ICE data sets. The identification accuracy of this method is 99.7, 98.13 and 99.26%, respectively, that is, higher than other methods in terms of both recognition accuracy and the number of used images.

Effect of conductance linearity and multi-level cell characteristics of TaOx-based synapse device on pattern recognition accuracy of neuromorphic system

To improve the classification accuracy of an image data set (CIFAR-10) by using analog input voltage, synapse devices with excellent conductance linearity (CL) and multi-level cell (MLC) characteristics are required. We analyze the CL and MLC characteristics of TaOx-based filamentary resistive random access memory (RRAM) to implement the synapse device in neural network hardware. Our findings show that the number of oxygen vacancies in the filament constriction region of the RRAM directly controls the CL and MLC characteristics. By adopting a Ta electrode (instead of Ti) and the hot-forming step, we could form a dense conductive filament. As a result, a wide range of conductance levels with CL is achieved and significantly improved image classification accuracy is confirmed.

Speech emotion recognition research: an analysis of research focus

This article analyses research in speech emotion recognition ("SER") from 2006 to 2017 in order to identify the current focus of research, and areas in which research is lacking. The objective is to examine what is being done in this field of research. Searching on selected keywords, we extracted and analysed 260 articles from well-known online databases. The analysis indicates that SER research is an active field of research, dozens of articles being published each year in journals and conference proceedings. The majority of articles concentrate on three critical aspects of SER, namely (1) databases, (2) suitable speech features, and (3) classification techniques to maximize the recognition accuracy of SER systems. Having carried out association analysis of the critical aspects and how they influence the performance of the SER system in term of recognition accuracy, we found that certain combination of databases, speech features and classifiers influence the recognition accuracy of the SER system. We have also suggested aspects of SER that could be taken into consideration in future works based on our review.

Compensated Synaptic Device for Improved Recognition Accuracy of Neuromorphic System

To improve a pattern recognition accuracy of synaptic device-based neuromorphic system, we tried to obtain symmetric conductance changes between conductance increase process (potentiation) and conductance decrease process (depression). By utilizing compensational voltage division, we achieved more gradual conductance changes during the depression, which led to the symmetric conductance changes between the potentiation and depression. On the basis of the achieved symmetric conductance changes, obviously improved pattern recognition accuracy was obtained on a multilayer perceptron structural simulation.

Developing Indian Sign Language Recognition System for Recognizing English Alphabets with Hybrid Classification Approach

Generally speaking and hearing are two simple practices of communication. the deaf and dumb people experiences difficult to communicate with normal people. To resolve this difficulty Indian Sign Language (ISL) recognition system is developed. our Indian Sign Language system uses both hand gesture image and NAM (Non-Audible-Murmur) speech to enhance accuracy of recognition system. From input image hand sign feature are taken by DWT (Discrete-Wavelet-Transform) after preprocessing. And from NAM speech features are taken by MFCC (Mel-Frequency-Cepstral-Coefficients). In this work we implemented fusing the image and audio features and as well as fusing the classification techniques for better recognition. We used HMM (Hidden-Markov-Model) and ANN (Artificial-Neural-Network) for classification. The experimental results shows maximum average recognition rate as 70.19% of the ISLR system while fusing sign image and NAM features with ANN classifier, 79.72% with HMM classifier and 88.84% while fusing classifiers.

A discriminant binarization transform using genetic algorithm and error-correcting output code for face template protection

In a biometric cryptosystem, biometric templates are encrypted before being stored in the database. This approach often requires a binarization phase to transform the original real-valued templates into their binary versions. Most of the existing binarization schemes do not take into consideration the need to discriminate binary templates; which in turn degrade the system recognition accuracy. This paper proposes a new binarization transform for face template protection using Error-Correcting Output Code (ECOC) and genetic algorithm (GA) which offsets this drawback. The proposed binarization transform novelty lies in the fact that the specific crossover, mutation and extension operators are defined by considering the theoretical properties of the ECOC framework in order to find an optimized coding matrix. The specification of these operations reduce the search space and speed up the convergence rate of GA. In addition, the proposed method is immune from being trapped in local minima. The quality of our method is evaluated using three well-known face databases: CMU PIE, Extended Yale B and FEI. The results demonstrate that the proposed binarization transform can result in a higher recognition accuracy without sacrificing the security of the protocol compared to alternative methods, such as binary discriminant analysis (BDA), discriminability-preserving (DP), chaos biohashing, and Discriminant-Genetic.

Segmentation- and Annotation-Free License Plate Recognition With Deep Localization and Failure Identification

Automated license plate recognition (ALPR) is essential in several roadway imaging applications. For ALPR systems deployed in the United States, variation between jurisdictions on character width, spacing, and the existence of noise sources (e.g., heavy shadows, non-uniform illumination, various optical geometries, poor contrast, and so on) present in LP images makes it challenging for the recognition accuracy and scalability of ALPR systems. Font and plate-layout variation across jurisdictions further adds to the difficulty of proper character segmentation and increases the level of manual annotation required for training classifiers for each state, which can result in excessive operational overhead and cost. In this paper, we propose a new ALPR workflow that includes novel methods for segmentation- and annotation-free ALPR, as well as improved plate localization and automation for failure identification. Our proposed workflow begins with localizing the LP region in the captured image using a two-stage approach that first extracts a set of candidate regions using a weak sparse network of winnows classifier and then filters them using a strong convolutional neural network (CNN) classifier in the second stage. Images that fail a primary confidence test for plate localization are further classified to identify localization failures, such as LP not present, LP too bright, LP too dark, or no vehicle found. In the localized plate region, we perform segmentation and optical character recognition (OCR) jointly by using a probabilistic inference method based on hidden Markov models (HMMs) where the most likely code sequence is determined by applying the Viterbi algorithm. In order to reduce manual annotation required for training classifiers for OCR, we propose the use of either artificially generated synthetic LP images or character samples acquired by trained ALPR systems already operating in other sites. The performance gap due to differences between training and target domain distributions is minimized using an unsupervised domain adaptation. We evaluated the performance of our proposed methods on LP images captured in several US jurisdictions under realistic conditions.

Robust coherence-based spectral enhancement for speech recognition in adverse real-world environments

Speech recognition in adverse real-world environments is highly affected by reverberation and non-stationary background noise. A well-known strategy to reduce such undesired signal components in multi-microphone scenarios is spatial filtering of the microphone signals. In this article, we demonstrate that an additional coherence-based postfilter, which is applied to the beamformer output signal to remove diffuse interference components from the latter, is an effective means to further improve the recognition accuracy of modern deep learning speech recognition systems. To this end, the 3rd CHiME Speech Separation and Recognition Challenge (CHiME-3) baseline speech enhancement system is extended by a coherence-based postfilter and the postfilter’s impact on the Word Error Rates (WERs) of a state-of-the-art automatic speech recognition system is investigated for the realistic noisy environments provided by CHiME-3. To determine the time- and frequency-dependent postfilter gains, we use Direction-of-Arrival (DOA)-dependent and (DOA)-independent estimators of the coherent-to-diffuse power ratio as an approximation of the short-time signal-to-noise ratio. Our experiments show that incorporating coherence-based postfiltering into the CHiME-3 baseline speech enhancement system leads to a significant reduction of the WERs, with relative improvements of up to 11.31%.

Speech recognition for under-resourced languages: Data sharing in hidden Markov model systems

&#x0D; &#x0D; &#x0D; For purposes of automated speech recognition in under-resourced environments, techniques used to share acoustic data between closely related or similar languages become important. Donor languages with abundant resources can potentially be used to increase the recognition accuracy of speech systems developed in the resource poor target language. The assumption is that adding more data will increase the robustness of the statistical estimations captured by the acoustic models. In this study we investigated data sharing between Afrikaans and Flemish – an under-resourced and well-resourced language, respectively. Our approach was focused on the exploration of model adaptation and refinement techniques associated with hidden Markov model based speech recognition systems to improve the benefit of sharing data. Specifically, we focused on the use of currently available techniques, some possible combinations and the exact utilisation of the techniques during the acoustic model development process. Our findings show that simply using normal approaches to adaptation and refinement does not result in any benefits when adding Flemish data to the Afrikaans training pool. The only observed improvement was achieved when developing acoustic models on all available data but estimating model refinements and adaptations on the target data only.&#x0D; &#x0D; &#x0D; &#x0D; &#x0D; Significance: &#x0D; &#x0D; &#x0D; &#x0D; Acoustic modelling for under-resourced languages&#x0D; Automatic speech recognition for Afrikaans&#x0D; Data sharing between Flemish and Afrikaans to improve acoustic modelling for Afrikaans&#x0D; &#x0D; &#x0D; &#x0D;

太赫兹安检系统人体图像边缘物体识别

An automatic recognition algorithm was proposed for recognizing objects concealed on the edge of human body in an image, which was scanned by THz human security system. First, the pretreatment methods, with binarization, filtering, filling, and morphological corrosion and expansion operations, were used to transfer the original THz image into a binary image, whose outline coordinates was later extracted by clockwise contour tracking algorithm. Then, the combinative application of circular template detection and non-minimal, non-maxima suppression algorithm were applied to calculate and select out all the convex and concave points in the extracted profile coordinates. Finally based on the grouping features of these adjacent convex and concave points, together with their distance constraints, the recognition of the edge objects(Objects covered the contour of human, etc.) could be realized. The experimental results, of testing 500 images of different people with different objects hiding on different edge position taking by the THz human security system, show that this algorithm represents quick identification and strong robustness, having the distinct advantages of strong anti-noise ability, high recognition speed and precision, and also it can match well the recognition accuracy of security system to achieve the system's resolution limit by adjusting proper parameters of the algorithm.

Face recognition Face2vec based on deep learning: Small database case

Object classification is a common problem in artificial intelligence and now it is usually approached by deep learning. In the paper the artificial neural network (ANN) architecture is considered. According to described ANN architecture, the ANN models are trained and tested on a relatively small Color-FERET facial image database under different conditions. The best fine-tuned ANN model provides 94% face recognition accuracy on Color-FERET frontal images and 98% face recognition accuracy within 3 attempts. However, for improving recognition system accuracy large data sets are still necessary preferably consisting of millions of images.

Accurate recognition of words in scenes without character segmentation using recurrent neural network

Recognition of texts in scenes is one of the most important tasks in many computer vision applications. Though different scene text recognition techniques have been developed, scene text recognition under a generic condition is still a very open and challenging research problem. One major factor that defers the advance in this research area is character touching, where many characters in scene images are heavily touched with each other and cannot be segmented for recognition. In this paper, we proposed a novel scene text recognition technique that performs word level recognition without character segmentation. Our proposed technique has three advantages. First it converts each word image into a sequential signal for the scene text recognition. Second, it adapts the recurrent neural network (RNN) with Long Short Term Memory (LSTM), the technique that has been widely used for handwriting recognition in recent years. Third, by integrating multiple RNNs, an accurate recognition system is developed which is capable of recognizing scene texts including those heavily touched ones without character segmentation. Extensive experiments have been conducted over a number of datasets including several ICDAR Robust Reading datasets and Google Street View dataset. Experiments show that the proposed technique is capable of recognizing texts in scenes accurately.

Recognition Algorithm and Optimization Experiments on Tomato Picking Robots

In order to improve the recognition accuracy of vision system on tomato picking robots, the paper proposed a method of feature extraction and recognition for ripe tomato based on illumination irrelevant images and support vector machine (SVM). In this method, we adopted vector median filter (VMF) to process the tomato images to eliminate noise and make the images more smooth firstly. To avoid the effects of natural environment illumination to the vision system, we processed tomato images and obtained the tomato illumination irrelevant images according to color constancy algorithm of the single pixel. Secondly, we segmented illumination irrelevant images using OSTU method, separated multiple objects by a watershed algorithm based on distance transform and got the target area with mathematical morphology. Also we extracted color, shape and textural features of the ripe tomatoes. Finally, we did experiments on recognizing tomatoes using support vector machine (SVM) with different kernel functions. At the same time, in order to obtain optimal model of SVM, we adopted cross validation and grid search method to optimize the model parameters. The experiment results show that illumination irrelevant processing not only can eliminate the influence of light intensity, but save a gray transferring step for further image segmentation. SVM with radial basis function is better than other kernel functions SVM and the tomato recognition accuracy is 95.7%. Through optimizing the parameter C and r of radial basis function, the tomato recognition accuracy reaches up to 96.9% with the increase of 1.2% when C and r is 4 and 16 respectively. This proves that it's feasible and effective to optimize SVM's parameters by cross validation and grid search method, which provide foundation for further research on vision system of tomato picking robots.

Lecturer performance system using neural network with Particle Swarm Optimization

ABSTRACTThe field of analyzing performance is very important and sensitive in particular when it is related to the performance of lecturers in academic institutions. Locating the weak points of lecturers through a system that provides an early warning to notify or reward the lecturers with warned or punished notices will help them to improve their weaknesses, leads to a better quality in the institutions. The current system has major issues in the higher education at Salahaddin University‐Erbil (SUE) in Kurdistan‐Iraq. These issues are: first, the assessment of lecturers’ activities is conducted traditionally via the Quality Assurance Teams at different departments and colleges at the university, second, the outcomes in some cases of lecturers’ performance provoke a low level of acceptance among lectures, as these cases are reflected and viewed by some academic communities as unfair cases, and finally, the current system is not accurate and vigorous. In this paper, Particle Swarm Optimization with Neural Network is used to assess performance of lecturers in more fruitful way and also to enhance the accuracy of recognition system. Different real and novel data sets are collected from SUE. The prepared datasets preprocessed and important features are then fed as input source to the training and testing phases. Particle Swarm Optimization is used to find the best weights and biases in the training phase of the neural network. The best accuracy rate obtained in the test phase is 98.28%. © 2016 Wiley Periodicals, Inc. Comput Appl Eng Educ 24:629–638, 2016; View this article online at wileyonlinelibrary.com/journal/cae; DOI 10.1002/cae.21737