Performance evaluation of pattern classifiers for handwritten character recognition

This paper describes a performance evaluation study in which some efficient classifiers are tested in handwritten digit recognition. The evaluated classifiers include a statistical classifier (modified quadratic discriminant function, MQDF), three neural classifiers, and an LVQ (learning vector quantization) classifier. They are efficient in that high accuracies can be achieved at moderate memory space and computation cost. The performance is measured in terms of classification accuracy, sensitivity to training sample size, ambiguity rejection, and outlier resistance. The outlier resistance of neural classifiers is enhanced by training with synthesized outlier data. The classifiers are tested on a large data set extracted from NIST SD19. As results, the test accuracies of the evaluated classifiers are comparable to or higher than those of the nearest neighbor (1-NN) rule and regularized discriminant analysis (RDA). It is shown that neural classifiers are more susceptible to small sample size than MQDF, although they yield higher accuracies on large sample size. As a neural classifier, the polynomial classifier (PC) gives the highest accuracy and performs best in ambiguity rejection. On the other hand, MQDF is superior in outlier rejection even though it is not trained with outlier data. The results indicate that pattern classifiers have complementary advantages and they should be appropriately combined to achieve higher performance. Received: July 18, 2001 / Accepted: September 28, 2001     

Recognition of handprinted numerals in VISA® card application forms

An optical character recognition (OCR) framework is developed and applied to handprinted numeric fields recognition. The numeric fields were extracted from binary images of VISA? credit card application forms. The images include personal identity numbers and telephone numbers. The proposed OCR framework is a cascaded neural networks. The first stage is a self-organizing feature map algorithm. The second stage maps distance values into allograph membership values using a gradient descent learning algorithm. The third stage is a multi-layer feedforward network. In this paper, we present experimental results which demonstrate the ability to read handprinted numeric fields. Experiments were performed on a test data set from the CCL/ITRI database which consists of over 90,390 handwritten numeric digits.     

Multiresolution recognition of unconstrained handwritten numerals with wavelet transform and multilayer cluster neural network

In this paper, we propose a new scheme for multiresolution recognition of unconstrained handwritten numerals using wavelet transform and a simple multilayer cluster neural network. The proposed scheme consists of two stages: a feature extraction stage for extracting multiresolution features with wavelet transform, and a classification stage for classifying unconstrained handwritten numerals with a simple multilayer cluster neural network. In order to verify the performance of the proposed scheme, experiments with unconstrained handwritten numeral database of Concordia University of Canada, Electro-Technical Laboratory of Japan, and Electronics and Telecommunications Research Institute of Korea were performed. The error rates were 3.20%, 0.83%, and 0.75%, respectively. These results showed that the proposed scheme is very robust in terms of various writing styles and sizes.     

基于Sugeno模糊积分神经网络分类器融合方法在手写数字识别中的应用

神经网络是模式识别中一种常见的分类器.针对同一个分类问题,构建多个分类器并把多个分类器进行融合可以提高分类系统的分类正确率、改善系统的稳健性.首先介绍了Sugeno模糊积分及Sugeno模糊积分神经网络分类器融合方法的一般原理,而后将其应用于手写数字识别,通过实际的案例验证了该融合方法的有效性和可行性.     

Merging polyhedral shapes with scattered features

1. initial embeddings of the polyhedra on unit spheres are computed, 2. the embeddings are deformed so that user-defined features (vertices) coincide on the spheres, and 3. an overlay of the subdivisions is computed and the aligned vertices are fused in the merged model.     

Handwritten numeral recognition based on simplified structural classification and fuzzy memberships

Previous handwritten numeral recognition algorithms applied structural classification to extract geometric primitives that characterize each image, and then utilized artificial intelligence methods, like neural network or fuzzy memberships, to classify the images. We propose a handwritten numeral recognition methodology based on simplified structural classification, by using a much smaller set of primitive types, and fuzzy memberships. More specifically, based on three kinds of feature points, we first extract five kinds of primitive segments for each image. A fuzzy membership function is then used to estimate the likelihood of these primitives being close to the two vertical boundaries of the image. Finally, a tree-like classifier based on the extracted feature points, primitives and fuzzy memberships is applied to classify the numerals. With our system, handwritten numerals in NIST Special Database 19 are recognized with correct rate between 87.33% and 88.72%.     

An architecture for handwritten text recognition systems

This paper presents an end-to-end system for reading handwritten page images. Five functional modules included in the system are introduced in this paper: (i) pre-processing, which concerns introducing an image representation for easy manipulation of large page images and image handling procedures using the image representation; (ii) line separation, concerning text line detection and extracting images of lines of text from a page image; (iii) word segmentation, which concerns locating word gaps and isolating words from a line of text image obtained efficiently and in an intelligent manner; (iv) word recognition, concerning handwritten word recognition algorithms; and (v) linguistic post-pro- cessing, which concerns the use of linguistic constraints to intelligently parse and recognize text. Key ideas employed in each functional module, which have been developed for dealing with the diversity of handwriting in its various aspects with a goal of system reliability and robustness, are described in this paper. Preliminary experiments show promising results in terms of speed and accuracy. Received October 30, 1998 / Revised January 15, 1999     

Recognition of handwritten musical notes by a modified Neocognitron

A neural network for recognition of handwritten musical notes, based on the well-known Neocognitron model, is described. The Neocognitron has been used for the what pathway (symbol recognition), while contextual knowledge has been applied for the where (symbol placement). This way, we benefit from dividing the process for dealing with this complicated recognition task. Also, different degrees of intrusiveness in learning have been incorporated in the same network: More intrusive supervised learning has been implemented in the lower neuron layers and less intrusive in the upper one. This way, the network adapts itself to the handwriting of the user. The network consists of a 13×49 input layer and three pairs of simple and complex neuron layers. It has been trained to recognize 20 symbols of unconnected notes on a musical staff and was tested with a set of unlearned input notes. Its recognition rate for the individual unseen notes was up to 93%, averaging 80% for all categories. These preliminary results indicate that a modified Neocognitron could be a good candidate for identification of handwritten musical notes.     

多尺度特征和神经网络相融合的手写体数字识别

针对传统特征提取方法无法有效解决书写随意性的干扰问题,提出了一种多尺度特征和神经网络相融合的手写体数字识别方法。首先提取手写体数字二值图像的轮廓、笔画次序等结构特征,并旋转坐标轴,提取多角度结构特征;然后将字符从中心点到外边框划分为K层矩形子层,提取每层图像的灰度特征,最后以两种多尺度特征构建神经网络模型,并预测测试集合样本。将该算法实际用于以MNIST字体库构建的两个数据集识别,其精度高达99.8%,并能有效降低倾斜等手写字体的随意性影响。     

Maximum mutual information training for an online neural predictive handwritten word recognition system

In this paper, we present a hybrid online handwriting recognition system based on hidden Markov models (HMMs). It is devoted to word recognition using large vocabularies. An adaptive segmentation of words into letters is integrated with recognition, and is at the heart of the training phase. A word-model is a left-right HMM in which each state is a predictive multilayer perceptron that performs local regression on the drawing (i.e., the written word) relying on a context of observations. A discriminative training paradigm related to maximum mutual information is used, and its potential is shown on a database of 9,781 words. Received June 19, 2000 / Revised October 16, 2000     

Rule-based document structure understanding with a fuzzy combination of layout and textual features

Document image processing is a crucial process in office automation and begins at the ‘OCR’ phase with difficulties in document ‘analysis’ and ‘understanding’. This paper presents a hybrid and comprehensive approach to document structure analysis. Hybrid in the sense that it makes use of layout (geometrical) as well as textual features of a given document. These features are the base for potential conditions which in turn are used to express fuzzy matched rules of an underlying rule base. Rules can be formulated based on features which might be observed within one specific layout object. However, rules can also express dependencies between different layout objects. In addition to its rule driven analysis, which allows an easy adaptation to specific domains with their specific logical objects, the system contains domain-independent markup algorithms for common objects (e.g., lists). Received June 19, 2000 / Revised November 8, 2000     

基于集成RBF神经网络的小类别手写体汉字识别系统

该文介绍了RBF神经网络的模型,讨论了RBF网络分类器的机理和特点,提出了一种集成RBF神经网络并应用于小类别手写体汉字识别系统的设计,采用了组合重心分解网格特征方法来提取汉字特征,设计了遗传进化隐层节点自生成算法用于RBF的训练。实验表明该小类别手写体汉字识别系统有很高的识别率,具有一定的实用推广价值。     

Recognition of handwritten Chinese characters by elastic matching

A method was proposed to match handwritten Chinese character patterns. Two given patterns are iteratively deformed until they match. An energy function and a neighborhood of influence is defined for each iteration. Initially a large neighborhood is used such that the movements result in large features being coarsely aligned. The neighborhood size is gradually reduced in successive iterations so that finer and finer details are aligned. The amount of computation increases with the square of the number of moving parts which is quite favorable compared with other algorithms. Extensive testing was carried out to evaluate the performance of the algorithm under various parameter settings. The method was applied to the recognition of handwritten Chinese characters with satisfactory results.     

一种量子神经网络模型及改进学习算法

提出一种量子BP网络模型及改进学习算法,该BP网络模型首先基于量子学中一位相移门和两位受控非门的通用性,构造出一种量子神经元,然后由该量子神经元构造隐含层,采用梯度下降法进行学习。输出层采用传统神经元构造,采用基于改进的带动量自适应学习率梯度下降法学习。在UCI两个数据集上采用该模型及算法,实验结果表明该方法比传统的BP网络具有较好的收敛速度和正确率。     

Experiments with adaptation strategies for a prototype-based recognition system for isolated handwritten characters

This paper describes an adaptive recognition system for isolated handwritten characters and the experiments carried out with it. The characters used in our experiments are alphanumeric characters, including both the upper- and lower-case versions of the Latin alphabets and three Scandinavian diacriticals. The writers are allowed to use their own natural style of writing. The recognition system is based on the k-nearest neighbor rule. The six character similarity measures applied by the system are all based on dynamic time warping. The aim of the first experiments is to choose the best combination of the simple preprocessing and normalization operations and the dissimilarity measure for a multi-writer system. However, the main focus of the work is on online adaptation. The purpose of the adaptations is to turn a writer-independent system into writer-dependent and increase recognition performance. The adaptation is carried out by modifying the prototype set of the classifier according to its recognition performance and the user's writing style. The ways of adaptation include: (1) adding new prototypes; (2) inactivating confusing prototypes; and (3) reshaping existing prototypes. The reshaping algorithm is based on the Learning Vector Quantization. Four different adaptation strategies, according to which the modifications of the prototype set are performed, have been studied both offline and online. Adaptation is carried out in a self-supervised fashion during normal use and thus remains unnoticed by the user. Received June 30, 1999 / Revised September 29, 2000     

A generic method of cleaning and enhancing handwritten data from business forms

The automation of business form processing is attracting intensive research interests due to its wide application and its reduction of the heavy workload due to manual processing. Preparing clean and clear images for the recognition engines is often taken for granted as a trivial task that requires little attention. In reality, handwritten data usually touch or cross the preprinted form frames and texts, creating tremendous problems for the recognition engines. In this paper, we contribute answers to two questions: “Why do we need cleaning and enhancement procedures in form processing systems?” and “How can we clean and enhance the hand-filled items with easy implementation and high processing speed?” Here, we propose a generic system including only cleaning and enhancing phases. In the cleaning phase, the system registers a template to the input form by aligning corresponding landmarks. A unified morphological scheme is proposed to remove the form frames and restore the broken handwriting from gray or binary images. When the handwriting is found touching or crossing preprinted texts, morphological operations based on statistical features are used to clean it. In applications where a black-and-white scanning mode is adopted, handwriting may contain broken or hollow strokes due to improper thresholding parameters. Therefore, we have designed a module to enhance the image quality based on morphological operations. Subjective and objective evaluations have been studied to show the effectiveness of the proposed procedures. Received January 19, 2000 / Revised March 20, 2001     

The recognition of handwritten numeral strings using a two-stage HMM-based method

In this paper, a two-stage HMM-based recognition method allows us to compensate for the possible loss in terms of recognition performance caused by the necessary trade-off between segmentation and recognition in an implicit segmentation-based strategy. The first stage consists of an implicit segmentation process that takes into account some contextual information to provide multiple segmentation-recognition hypotheses for a given preprocessed string. These hypotheses are verified and re-ranked in a second stage by using an isolated digit classifier. This method enables the use of two sets of features and numeral models: one taking into account both the segmentation and recognition aspects in an implicit segmentation-based strategy, and the other considering just the recognition aspects of isolated digits. These two stages have been shown to be complementary, in the sense that the verification stage compensates for the loss in terms of recognition performance brought about by the necessary tradeoff between segmentation and recognition carried out in the first stage. The experiments on 12,802 handwritten numeral strings of different lengths have shown that the use of a two-stage recognition strategy is a promising idea. The verification stage brought about an average improvement of 9.9% on the string recognition rates. On touching digit pairs, the method achieved a recognition rate of 89.6%. Received June 28, 2002 / Revised July 03, 2002     

A criterion based on Fourier transform for segmentation of connected digits

Abstract. Segmentation is the most difficult problem in handwritten character recognition systems and often causes major errors in performance. To reach a balance between speed and accuracy, a filter distinguishing connected images from isolated images for multiple stage segmentation is required. The Fourier spectrum is a promising approach to this problem, although it suffers from the heavy influence of stroke width. Therefore, we introduce SFS (SFS) to eliminate the stroke-width effect. Based on the SFS, a set of features and a fine-tuned criterion are presented to classify connected/isolated images. Theoretical analysis demonstrates their soundness, while experimental results demonstrate that this criterion is better than other methods. Received February 18, 2000 / Revised June 3, 2000     

Neocognitron with improved bend-extractors: Recognition of handwritten digits in the real world

We have reported previously that the performance of a neocognitron can be improved by a built-in bend-extracting layer. The conventional bend-extracting layer can detect bend points and end points of lines correctly, but not always crossing points of lines. This paper shows that an introduction of a mechanism of disinhibition can make the bend-extracting layer detect not only bend points and end points, but also crossing points of lines correctly. This paper also demonstrates that a neocognitron with this improved bend-extracting layer can recognise handwritten digits in the real world with a recognition rate of about 98%. We use the technique of dual thresholds for feature-extracting S-cells, and higher threshold values are used in the learning than in the recognition phase. We discuss how the threshold values affect the recognition rate.     

Reduction of expanded search terms for fuzzy English-text retrieval

Optical character reader (OCR) misrecognition is a serious problem when OCR-recognized text is used for retrieval purposes in digital libraries. We have proposed fuzzy retrieval methods that, instead of correcting the errors manually, assume that errors remain in the recognized text. Costs are thereby reduced. The proposed methods generate multiple search terms for each input query term by referring to confusion matrices, which store all characters likely to be misrecognized and the respective probability of each misrecognition. The proposed methods can improve recall rates without decreasing precision rates. However, a few million search terms are occasionally generated in English-text fuzzy retrieval, giving an intolerable effect on retrieval speed. Therefore, this paper presents two remedies to reduce the number of generated search terms while maintaining retrieval effectiveness. One remedy is to restrict the number of errors included in each expanded search term, while the other is to introduce another validity value different to our conventional one. Experimental results indicate that the former remedy reduced the number of terms to about 50 and the latter to not more than 20. Received: 18 December 1998 / Revised: 31 May 1999