Recognizing chinese characters through interactive activation and competition

Most conventional approaches to Chinese character recognition attempt to recognize unknown Chinese characters per se. Hence the discriminative power of features employed play a crucial role in determining the ultimate recognition rate. However, high level linguistic knowledge such as the strong word context effects present in the Chinese language and the different frequencies of words in the daily use of the language, are not utilized in machine recognition of Chinese characters. In this paper, a word-oriented recognizer using the Interactive Activation and Competition model (IAC model) is proposed. Such a recognizer is tolerant to noise, size and font variations and it also possesses self-learning capability.     

Sichuan dialect speech recognition with deep LSTM network

In speech recognition research,because of the variety of languages,corresponding speech recognition systems need to be constructed for different languages.Especially in a dialect speech recognition system,there are many special words and oral language features.In addition,dialect speech data is very scarce.Therefore,constructing a dialect speech recognition system is difficult.This paper constructs a speech recognition system for Sichuan dialect by combining a hidden Markov model(HMM)and a deep long short-term memory(LSTM)network.Using the HMM-LSTM architecture,we created a Sichuan dialect dataset and implemented a speech recognition system for this dataset.Compared with the deep neural network(DNN),the LSTM network can overcome the problem that the DNN only captures the context of a fixed number of information items.Moreover,to identify polyphone and special pronunciation vocabularies in Sichuan dialect accurately,we collect all the characters in the dataset and their common phoneme sequences to form a lexicon.Finally,this system yields a 11.34%character error rate on the Sichuan dialect evaluation dataset.As far as we know,it is the best performance for this corpus at present.     

A SVM-based cursive character recognizer

This paper presents a cursive character recognizer, a crucial module in any cursive word recognition system based on a segmentation and recognition approach. The character classification is achieved by using support vector machines(SVMs) and a neural gas. The neural gas is used to verify whether lower and upper case version of a certain letter can be joined in a single class or not. Once this is done for every letter, the character recognition is performed by SVMs. A database of 57 293 characters was used to train and test the cursive character recognizer. SVMs compare notably better, in terms of recognition rates, with popular neural classifiers, such as learning vector quantization and multi-layer-perceptron. SVM recognition rate is among the highest presented in the literature for cursive character recognition.     

神经网络在车辆牌照字符识别中的应用

在车辆牌照自动识别系统中,因自然因素或采用因素使得原本原则的印刷字符产生畸变,给字符识别带来了很大困难。本文在特征抽取的基础上,采用ＢＰ网络进行分类,并附加线性感知器来实现单字的有效识别。该方法算法简便,识别率高,可适用于多种高噪声环境中的印刷字体识别。     

一个基于BP网络的字符识别器

本文简单介绍了人工神经元网络的背景知识,提出了一种利用传统BP（Back-Propagation误差逆传播）网络识别印刷字符的方法,用C语言对其进行了实现．在进行了大量实验之后,实验结果表明该字符识别器具有较好的有效性和正确性,能够在合理的误差范围内以较好的效率成功识别字符．     

A new representation of shape and its use for high performance in online Arabic character recognition by an associative memory

The purpose of this study is to investigate a new representation of shape and its use in handwritten online character recognition by a Kohonen associative memory. This representation is based on the empirical distribution of features such as tangents and tangent differences at regularly spaced points along the character signal. Recognition is carried out by a Kohonen neural network trained using the representation. In addition to the Euclidean distance traditionally used in the Kohonen training algorithm to measure the similarities among feature vectors, we also investigate the Kullback–Leibler divergence and the Hellinger distance, functions that measure distance between distributions. Furthermore, we perform operations (pruning and filtering) on the trained memory to improve its classification potency. We report on extensive experiments using a database of online Arabic characters produced without constraints by a large number of writers. Comparative results show the pertinence of the representation and the superior performance of the scheme.     

Pair-wise discrimination based on a stroke importance measure

The pair-wise discriminator is a binary classifier that verifies the outcome of the recognizer if it belongs to a class in a pre-defined confusion pair database. It is difficult to discriminate a pair of characters that are very similar in shape except for a small difference, because the small difference can be overridden by the writing variation. This paper proposes a pair-wise discrimination method that discriminates similar characters by focusing on the structural difference between the two characters. It discriminates a pair of characters by comparing their matching scores between the input character and the models of the two characters. When the stroke matching scores are combined to compute the overall matching score, each stroke is assigned a weight to reflect its importance in discriminating the character pair. By assigning large weights to the discriminative strokes, the difference between the characters is emphasized. The stroke weights are systematically obtained by a neural network training algorithm. In the experiments, the recognition performance was significantly improved by applying the proposed method.     

基于TDNN-FSMN的蒙古语语音识别技术研究

为了提高蒙古语语音识别性能,该文首先将时延神经网络融合前馈型序列记忆网络应用于蒙古语语音识别任务中,通过对长序列语音帧建模来充分挖掘上下文相关信息;此外研究了前馈型序列记忆网络“记忆”模块中历史信息和未来信息长度对模型的影响;最后分析了融合的网络结构中隐藏层个数及隐藏层节点数对声学模型性能的影响。实验结果表明,时延神经网络融合前馈型序列记忆网络相比深度神经网络、时延神经网络和前馈型序列记忆网络具有更好的性能,单词错误率与基线深度神经网络模型相比降低22.2%。     

Resolution of pattern recognition problems using a hybrid Genetic/Random Neural Network learning algorithm

Gelenbe has proposed a neural network, called a Random Neural Network, which calculates the probability of activation of the neurons in the network. In this paper, we propose to solve the patterns recognition problem using a hybrid Genetic/Random Neural Network learning algorithm. The hybrid algorithm trains the Random Neural Network by integrating a genetic algorithm with the gradient descent rule-based learning algorithm of the Random Neural Network. This hybrid learning algorithm optimises the Random Neural Network on the basis of its topology and its weights distribution. We apply the hybrid Genetic/Random Neural Network learning algorithm to two pattern recognition problems. The first one recognises or categorises alphabetic characters, and the second recognises geometric figures. We show that this model can efficiently work as associative memory. We can recognise pattern arbitrary images with this algorithm, but the processing time increases rapidly.     

Writer adaptation via deeply learned features for online Chinese handwriting recognition

This paper proposes a novel framework of writer adaptation based on deeply learned features for online handwritten Chinese character recognition. Our motivation is to further boost the state-of-the-art deep learning-based recognizer by using writer adaptation techniques. First, to perform an effective and flexible writer adaptation, we propose a tandem architecture design for the feature extraction and classification. Specifically, a deep neural network (DNN) or convolutional neural network (CNN) is adopted to extract the deeply learned features which are used to build a discriminatively trained prototype-based classifier initialized by Linde–Buzo–Gray clustering techniques. In this way, the feature extractor can fully utilize the useful information of a DNN or CNN. Meanwhile, the prototype-based classifier could be designed more compact and efficient as a practical solution. Second, the writer adaption is performed via a linear transformation of the deeply learned features which is optimized with a sample separation margin-based minimum classification error criterion. Furthermore, we improve the generalization capability of the previously proposed discriminative linear regression approach for writer adaptation by using the linear interpolation of two transformations and adaptation data perturbation. The experiments on the tasks of both the CASIA-OLHWDB benchmark and an in-house corpus with a vocabulary of 20,936 characters demonstrate the effectiveness of our proposed approach.     

基于Viterbi 算法的粘连断裂印刷体数字行切分识别方法

粘连断裂字符行的切分识别，是很多OCR 实际应用中存在的主要困难之一. 本文针对粘连断裂的印刷体数字行，提出了一种基于Viterbi 算法的切分识别方案，该方案采用两次切分识别的层次型结构. 在第二次切分识别过程中，首先，在候选切分点区域，结合灰度图像与二值轮廓信息，采用基于Viterbi 算法搜索的非直线路径进行切分，得到有效的切分路径；然后，结合分类器输出的可信度，采用Viterbi 算法来合并前面得到的候选切分图像块，进行动态切分与识别. 实际的金融票据识别系统实验表明，本文提出的印刷体数字行切分识别方法能够较好的克服字符行的粘连与断裂情况，提高了识别系统的识别率和鲁棒性.     

Modified region decomposition method and optimal depth decision tree in the recognition of non-uniform sized characters – An experimentation with Kannada characters

In contrast to English alphabets, some characters in Indian languages such as Kannada, Hindi, Telugu may have either horizontal or vertical or both the extensions making it difficult to enclose every such character in a standard rectangular grid as done quite often in character recognition research. In this work, an improved method is proposed for the recognition of such characters (especially Kannada characters), which can have spread in vertical and horizontal directions. The method uses a standard sized rectangle which can circumscribe standard sized characters. This rectangle can be interpreted as a two-dimensional, 3×3 structure of nine parts which we define as bricks. This structure is also interpreted as consecutively placed three row structures of three bricks each or adjacently placed three column structures of three bricks each.

It is obvious that non-uniform sized characters cannot be contained within the standard rectangle of nine bricks. The work presented here proposes to take up such cases. If the character has horizontal extension, then the rectangle is extended horizontally by adding one column structure of three bricks at a time, until the character is encapsulated. Likewise, for vertically extended characters, one row structure is added at a time. For the characters which are smaller than the standard rectangle, one column structure is removed at a time till the character fits in the shrunk rectangle. Thus, the character is enclosed in a rectangular structure of m×n bricks where m3 and n1. The recognition is carried out intelligently by examining certain selected bricks only instead of all mn bricks. The recognition is done based on an optimal depth logical decision tree developed during the Learning phase and does not require any mathematical computation.     

基于DNN处理的鲁棒性I-Vector说话人识别算法

提出了一种将基于深度神经网络（Deep Neural Network,DNN）特征映射的回归分析模型应用到身份认证矢量（identity vector,i-vector）/概率线性判别分析（Probabilistic Linear Discriminant Analysis,PLDA）说话人系统模型中的方法。DNN通过拟合含噪语音和纯净语音i-vector之间的非线性函数关系,得到纯净语音i-vector的近似表征,达到降低噪声对系统性能影响的目的。在TIMIT数据集上的实验验证了该方法的可行性和有效性。     

基于DNN与基音周期的说话人识别

传统说话人识别框架大多建立在高斯混合模型（GMM）上的,然而这种浅层学习模型不能有效地表征数据特征之间的高阶相关性,识别效果较差。本文提出一种基于深度神经网络（Deep Neural Network, DNN）与基音周期（Pitch Period, PP）相结合的说话人识别方法,模型主线识别以对数梅尔滤波器组特征参数作为DNN的输入,通过训练DNN模型提取说话人的声纹特征;针对DNN模型阈值设定人的主观性影响,利用动态时间规整技术匹配说话人基音周期进行辅助识别。实验结果表明,这种双重识别方法等错误率可以达到1.6%,较DNN系统与EM-GMM系统等错误率分别降低了1.2%和2.4%,并且在噪声环境中仍具有较好的鲁棒性。     

Non-uniform cellular automata based associative memory: Evolutionary design and basins of attraction

This paper presents the synthesis and analysis of a special class of non-uniform cellular automata (CAs) based associative memory, termed as generalized multiple attractor CAs (GMACAs). A reverse engineering technique is presented for synthesis of the GMACAs. The desired CAs are evolved through an efficient formulation of genetic algorithm coupled with the reverse engineering technique. This has resulted in significant reduction of the search space of the desired GMACAs. Characterization of the basins of attraction of the proposed model establishes the sparse network of GMACAs as a powerful pattern recognizer for memorizing unbiased patterns. Theoretical analysis also provides an estimate of the noise accommodating capability of the proposed GMACA based associative memory. An in-depth analysis of the GMACA rule space establishes the fact that more heterogeneous CA rules are capable of executing complex computation like pattern recognition.     

Handwritten Chinese text editing and recognition system

This paper describes a handwritten Chinese text editing and recognition system that can edit handwritten text and recognize it with a client-server mode. First, the client end samples and redisplays the handwritten text by using digital ink technics, segments handwritten characters, edits them and saves original handwritten information into a self-defined document. The self-defined document saves coordinates of all sampled points of handwriting characters. Second, the server recognizes handwritten document based on the proposed Gabor feature extraction and affinity propagation clustering (GFAP) method, and returns the recognition results to client end. Moreover, the server can also collect the labeled handwritten characters and fine tune the recognizer automatically. Experimental results on HIT-OR3C database show that our handwriting recognition method improves the recognition performance remarkably.     

Handwritten digit recognition by means of a holographic associative memory

In this paper, a holographic associative memory (HAM) is proposed for recognizing handwritten variations of the ten digits. First, the handwritten characters were taken from the NIST standard database in order to extract relevant features from each one of them. Each digit was thus represented as a vector of 112 features constructed by dividing each character into 16 equal-sized partitions, each one used to extract seven different features for recognition. Second, these feature vectors, and reduced combinations of them, were input to train several HAM systems respectively. Then, all these memories were tested with a new set of patterns and the lowest-error HAM was chosen as the best training set. The features used in this last memory were taken as the most significant variables for describing each digit in the database. Finally, these most significant features were used to show the behaviour of the recognition rate when training the HAM with reduced training sets. Some final conclusions are reported and future work directions are proposed.