A new methodology for gray-scale character segmentation andrecognition

Generally speaking, through the binarization of gray-scale images, useful information for the segmentation of touched or overlapped characters may be lost in many cases. If we analyze gray-scale images, however, specific topographic features and the variation of intensities can be observed in the character boundaries. In this paper, we propose a new methodology for character segmentation and recognition which makes the best use of the characteristics of gray-scale images. In the proposed methodology, the character segmentation regions are determined by using projection profiles and topographic features extracted from the gray-scale images. Then a nonlinear character segmentation path in each character segmentation region is found by using multi-stage graph search algorithm. Finally, in order to confirm the nonlinear character segmentation paths and recognition results, a recognition-based segmentation method is adopted. Through the experiments with various kinds of printed documents, it is convinced that the proposed methodology is very effective for the segmentation and recognition of touched and overlapped characters     

POSIT: Part-based object segmentation without intensive training

Object segmentation is a well-known difficult problem in pattern recognition. Until now, most of the existing object segmentation methods need to go through a time-consuming training phase prior to segmentation. Both robustness and efficiency of the existing methods have room for improvement. In this work, we propose a new methodology, called POSIT, for object segmentation without intensive training process. We construct a part-based shape model to substitute the training process. In the part-based framework, we sequentially register object parts in the prior model to an image so that the searching space is largely reduced. Another advantage of the sequential matching is that, instead of predefining the weighting parameters for the terms in the matching evaluation function, we can estimate the parameters in our model on the fly. Finally, we fine-tune the previous coarse segmentation by localized graph cuts. In the experiments, POSIT has been tested on numerous natural horse and cow images and the obtained results show the accuracy, robustness and efficiency of the proposed object segmentation method.     

Optical character recognition for cursive handwriting

A new analytic scheme, which uses a sequence of image segmentation and recognition algorithms, is proposed for the off-line cursive handwriting recognition problem. First, some global parameters, such as slant angle, baselines, stroke width and height, are estimated. Second, a segmentation method finds character segmentation paths by combining gray-scale and binary information. Third, a hidden Markov model (HMM) is employed for shape recognition to label and rank the character candidates. For this purpose, a string of codes is extracted from each segment to represent the character candidates. The estimation of feature space parameters is embedded in the HMM training stage together with the estimation of the HMM model parameters. Finally, information from a lexicon and from the HMM ranks is combined in a graph optimization problem for word-level recognition. This method corrects most of the errors produced by the segmentation and HMM ranking stages by maximizing an information measure in an efficient graph search algorithm. The experiments indicate higher recognition rates compared to the available methods reported in the literature     

Real-time arabic handwritten character recognition

This paper considers the development of a real-time Arabic handwritten character recognition system. The shape of an Arabic character depends on its position in a given word. The system assumes that characters result from a reliable segmentation stage, thus, the position of the character is known a priori. Thus, four different sets of character shapes have been independently considered. Each set is further divided into four subsets depending on the number of strokes in the character. The system has been heavily tested and the average recognition rate has been found to be 99.6% where most of the misrecognized characters were actually written with little care. Thus, the system can be reliably used for the recognition of on-line handwritten characters entered via a graphic tablet.     

A Character Flow Framework for Multi-Oriented Scene Text Detection

Scene text detection plays a significant role in various applications,such as object recognition,document management,and visual navigation.The instance segmentation based method has been mostly used in existing research due to its advantages in dealing with multi-oriented texts.However,a large number of non-text pixels exist in the labels during the model training,leading to text mis-segmentation.In this paper,we propose a novel multi-oriented scene text detection framework,which includes two main modules:character instance segmentation (one instance corresponds to one character),and character flow construction (one character flow corresponds to one word).We use feature pyramid network(FPN) to predict character and non-character instances with arbitrary directions.A joint network of FPN and bidirectional long short-term memory (BLSTM) is developed to explore the context information among isolated characters,which are finally grouped into character flows.Extensive experiments are conducted on ICDAR2013,ICDAR2015,MSRA-TD500 and MLT datasets to demonstrate the effectiveness of our approach.The F-measures are 92.62％,88.02％,83.69％ and 77.81％,respectively.     

Facial deblur inference using subspace analysis for recognition of blurred faces

This paper proposes a novel method for recognizing faces degraded by blur using deblurring of facial images. The main issue is how to infer a Point Spread Function (PSF) representing the process of blur on faces. Inferring a PSF from a single facial image is an ill-posed problem. Our method uses learned prior information derived from a training set of blurred faces to make the problem more tractable. We construct a feature space such that blurred faces degraded by the same PSF are similar to one another. We learn statistical models that represent prior knowledge of predefined PSF sets in this feature space. A query image of unknown blur is compared with each model and the closest one is selected for PSF inference. The query image is deblurred using the PSF corresponding to that model and is thus ready for recognition. Experiments on a large face database (FERET) artificially degraded by focus or motion blur show that our method substantially improves the recognition performance compared to existing methods. We also demonstrate improved performance on real blurred images on the FRGC 1.0 face database. Furthermore, we show and explain how combining the proposed facial deblur inference with the local phase quantization (LPQ) method can further enhance the performance.     

Segmentation of cardiac tagged MR images using a snake model based on hybrid gradient vector flow

In the segmentation of cardiac tagging magnetic resonance (tMR) images, it is difficult to segment the left ventricle automatically by using the traditional segmentation model because of the interference caused by the tags. A new snake model based on hybrid gradient vector flow (HGVF) is proposed by us to improve this segmentation. Due to the different characteristics between endocardium and epicardium of the left ventricle (LV), several gradient vector flows (GVFs) with distinctive boundary information would be fused to segment these two sub regions individually. For segmentation of endocardium, we construct a new HGVF in snake model fused by three independent GVFs. These flows are respectively exported from the original cardiac tMR image, the tags-removed image and the local-filtered image. On the other hand, since the epicardium is with a nearly-circle shape, we construct the other HGVF which is composed of two different GVFs. One of them is derived from the tags-removed image either and the other one is derived from the ideal circle-shape image. Some experiments have been done to validate our new segmentation model. The average overlap of the endocardium segmentation is 89.67% (its mean absolute distance is 1.86 pixels), and the average overlap of the epicardium segmentation is 95.88% (its mean absolute distance is 1.64 pixels). Experimental results show that the proposed method improves the segmentation performance compared to some available methods effectively.     

车牌识别中先验知识的嵌入及字符分割方法

针对车牌字符分割过程中先验知识嵌入困难,分割过程对于前期车牌定位依赖较强的问题,提出了一种新的先验知识嵌入方法及其对应的字符分割算法。给定一种类型的车牌,利用字符的可能排列方式定义马尔可夫链中的状态,可以将车牌字符分割转化为一组马尔可夫链的前向识别过程。结合连通分量提取及垂直投影分割算法,可以有效地获取车牌的最优分割结果及其置信度。在实际应用中,该算法不依赖于前期的精确定位,对粗定位后的图像即可进行快速有效地分割。该方法统一了不同类型车牌的先验知识嵌入方法,降低了编码复杂度。在中国车牌及马来西亚车牌上的实验结果均证明,该方法有效地提高了车牌字符分割的性能。     

Building A Highly Accurate Mandarin Speech Recognizer With Language-Independent Technologies and Language-Dependent Modules

We describe a system for highly accurate large-vocabulary Mandarin speech recognition. The prevailing hidden Markov model based technologies are essentially language independent and constitute the backbone of our system. These include minimum-phone-error discriminative training and maximum-likelihood linear regression adaptation, among others. Additionally, careful considerations are taken into account for Mandarin-specific issues including lexical word segmentation, tone modeling, phone set design, and automatic acoustic segmentation. Our system comprises two sets of acoustic models for the purposes of cross adaptation. The systems are designed to be complementary in terms of errors but with similar overall accuracy by using different phone sets and different combinations of discriminative learning. The outputs of the two subsystems are then rescored by an adapted n-gram language model. Final confusion network combination yielded 9.1% character error rate on the DARPA GALE 2007 official evaluation, the best Mandarin recognition system in that year.     

Scene word recognition from pieces to whole

Convolutional neural networks (CNNs) have had great success with regard to the object classification problem. For character classification, we found that training and testing using accurately segmented character regions with CNNs resulted in higher accuracy than when roughly segmented regions were used. Therefore, we expect to extract complete character regions from scene images. Text in natural scene images has an obvious contrast with its attachments. Many methods attempt to extract characters through different segmentation techniques. However, for blurred, occluded, and complex background cases, those methods may result in adjoined or over segmented characters. In this paper, we propose a scene word recognition model that integrates words from small pieces to entire after-cluster-based segmentation. The segmented connected components are classified as four types: background, individual character proposals, adjoined characters, and stroke proposals. Individual character proposals are directly inputted to a CNN that is trained using accurately segmented character images. The sliding window strategy is applied to adjoined character regions. Stroke proposals are considered as fragments of entire characters whose locations are estimated by a stroke spatial distribution system. Then, the estimated characters from adjoined characters and stroke proposals are classified by a CNN that is trained on roughly segmented character images. Finally, a lexicondriven integration method is performed to obtain the final word recognition results. Compared to other word recognition methods, our method achieves a comparable performance on Street View Text and the ICDAR 2003 and ICDAR 2013 benchmark databases. Moreover, our method can deal with recognizing text images of occlusion and improperly segmented text images.     

标牌粘连字符自适应定位分割重建与识别

目的 针对仪表、电梯等标牌上一些字符间距较小,传统分割方法分割不准确,字符识别率不高的问题,提出了一种标牌粘连字符自适应定位分割重建识别算法。方法 首先对标牌图像进行中值滤波、二值化等预处理;其次运用数学形态学方法对预处理后的图像进行开运算及腐蚀,将字符间一些无用的信息去掉,增大字符间距;继而通过形心算法找出每个字符的几何中心,并通过Sobel边缘检测算子根据几何中心获取每个字符边框,建立ROI（region of interest）,再返回标牌原图利用已经建立的ROI从中分割字符,依据国家字符间距相关标准,在分割的每个字符后加一定像素宽的矩形间隔条后重建字符图像,再进行OCR（optical character recognition）字符识别。结果 经过对993块标牌进行字符识别实验,算法的识别率达到95.7%。结论 实验结果表明本文算法是对标牌字符识别的一种有效算法。     

A novel adaptive morphological approach for degraded character image segmentation

This work proposes a novel adaptive approach for character segmentation and feature vector extraction from seriously degraded images. An algorithm based on the histogram automatically detects fragments and merges these fragments before segmenting the fragmented characters. A morphological thickening algorithm automatically locates reference lines for separating the overlapped characters. A morphological thinning algorithm and the segmentation cost calculation automatically determine the baseline for segmenting the connected characters. Basically, our approach can detect fragmented, overlapped, or connected character and adaptively apply for one of three algorithms without manual fine-tuning. Seriously degraded images as license plate images taken from real world are used in the experiments to evaluate the robustness, the flexibility and the effectiveness of our approach. The system approach output data as feature vectors keep useful information more accurately to be used as input data in an automatic pattern recognition system.     

低秩和主动形状模型的肺区分割

目的 肺区分割是肺癌计算机辅助诊断系统的首要步骤。主动形状模型（active shape model,ASM）能根据训练集获得肺区形状模型,再结合待分割肺区影像自身的局部特征,进行测试影像的分割。由于主成分分析（principal component analysis,PCA）仅能去除服从高斯分布的噪声,不能处理其他类型的噪声,所以当训练集含有非高斯类型的噪声样本时,采用基于PCA的ASM无法训练出正确的形状模型,使得肺区分割不能得到正确的结果。而低秩（low rank,LR）理论的鲁棒主成分分析（robust principal component analysis,RPCA）能去除各种类型的噪声,基于此,本文提出一种将RPCA与ASM相结合的方法。方法 首先对训练样本集标记点矩阵进行低秩分解,去除噪声样本对训练出的形状模型的影响。然后在ASM训练局部梯度模型时,用判断训练样本轮廓上的标记点曲率直方图的相似度来去除噪声样本。结果 在训练集含噪声样本时,将基于RPCA的ASM与传统ASM（即基于PCA的ASM）分别生成的形状模型进行对比,发现基于RPCA的ASM生成的形状模型与训练集无噪声样本时传统ASM生成的形状模型更相符。在训练集含噪声样本的情况下,基于RPCA的ASM方法分割EMPIRE10数据集中的22个肺影像,与金标准的重叠度为94.5%,而基于PCA的ASM方法分割准确率仅为69.5%。结论 实验结果表明,在训练样本集中有噪声样本的情况下,基于RPCA的ASM分割能得到比基于PCA的ASM更好的分割效果。     

Time-segmentation and position-free recognition of air-drawn gestures and characters in videos

We report the recognition in video streams of isolated alphabetic characters and connected cursive textual characters, such as alphabetic, hiragana and kanji characters, that are drawn in the air. This topic involves a number of difficult problems in computer vision, such as the segmentation and recognition of complex motion on videos. We use an algorithm called time–space continuous dynamic programming (TSCDP), which can realize both time- and location-free (spotting) recognition. Spotting means that the prior segmentation of input video is not required. Each reference (model) character is represented by a single stroke that is composed of pixels. We conducted two experiments involving the recognition of 26 isolated alphabetic characters and 23 Japanese hiragana and kanji air-drawn characters. We also conducted gesture recognition experiments based on TSCDP, which showed that TSCDP was free from many of the restrictions imposed by conventional methods.     

基于小波和DCT的灰度压印字符图像的特征抽取

标牌压印字符是“反光差”的凹凸字符,通常的基于二值化图像的字符特征抽取方法都不适宜。提出了基于灰度图像的标牌压印字符特征抽取新方法,首先对灰度字符进行圆周投影,然后利用小波变换,将投影曲线分解为大致信号和细节信号的子样本,最后对子样本进行DCT变换,生成凹凸字符的特征矢量。该方法是直接对灰度图像抽取字符特征,不仅可以尽量多地保持原始字符的特征,而且克服了传统的字符图像特征抽取时,过分依赖于二值化算法、抗干扰性差等弊病。对标牌压印有限凹凸字符集进行特征抽取和识别实验表明,该特征抽取方法具有尺度和旋转不变性,有较好的抗干扰性和很好的分类性能,实用价值很高。     

Handwritten Chinese character recognition using nonlinear active shape models and the Viterbi algorithm

Since Chinese characters are composed from a small set of fundamental shapes (radicals) the problem of recognising large numbers of characters can be converted to that of extracting a small number of radicals and then finding their optimal combination. In this paper, radical extraction is carried out by nonlinear active shape models, in which kernel principal component analysis is employed to capture the nonlinear variation. Treating Chinese character composition as a discrete Markov process, we also propose an approach to recognition with the Viterbi algorithm. Our initial experiments are conducted on off-line recognition of 430,800 loosely-constrained characters, comprised of 200 radical categories covering 2154 character categories from 200 writers. The correct recognition rate is 93.5% characters correct (writer-independent). Consideration of published figures for existing radical approaches suggests that our method achieves superior performance.     

基于字符分割和LeNet-5网络的字符验证码识别

为了解决传统验证码识别方法效率低,精度差的问题,设计了一种先分割后识别的验证码处理方案。该方案在预处理阶段用中值滤波去噪,再利用霍夫变换对图像字符进行矫正;在字符分割阶段,利用垂直投影算法确定验证码字符块个数,以及字符坐标点,再用颜色填充算法对验证码进行初步分割,根据分割后的字符块数量对粘连字符进行二次分割;在识别阶段,我们对LeNet-5网络进行了改进,修改了输入层,并用全连接层替换了LeNet-5网络中的C5层,以此来对验证码字符进行识别;实验表明,对于非粘连验证码和粘连验证码,单张图片分割时间为0.14和0.15ms,分割准确率为98.75%和97.25%,识别准确率为99.99%和97.7%;结果表明,该算法对验证码分割和识别都有着很好的效果。