2DVTE: A two-directional videotext extractor for rapid and elaborate design

In video indexing and summarization, videotext is the very compact and accurate information. Most videotext detection and extraction methods only deal with the static videotext on video frames. Few methods can handle motion videotext efficiently since motion videotext is hardly extracted well. In this paper, we propose a two-directional videotext extractor, called 2DVTE. It is developed as an integrated system to detect, localize and extract the scrolling videotexts. First, the detection method is carried out by edge information to classify regions into text and non-text regions. Second, referring to the localization on scrolling videotext, we propose the two-dimensional projection profile method with horizontal and vertical edge map information. Considering the characteristics of Chinese text, the vertical edge map is used to localize the possible text region and horizontal edge map is used to refine the text region. Third, the extraction method consists of dual mode adaptive thresholding and multi-seed filling algorithm. In the dual mode adaptive thresholding, it produces the non-rectangle pattern to divide the background and foreground more precisely. Referring to the multi-seed filling algorithm, it is based on the consideration of the minimum and maximum length and four directions of the stroke while the previous method only considers the minimum length and two directions of the stroke. With this multi-seed exploitation on strokes, precise seeds are obtained to produce more sophisticated videotext. Considering high throughput and the low complexity issue, we can achieve a real-time system on detecting, localizing, and extracting the scrolling videotexts with only one frame usage instead of multi-frame integration in other literatures. According to the experiment results on various video sequences, all of the horizontal and vertical scrolling videotexts can be extracted precisely. We also make comparisons with other methods. In our analysis, the performance of our algorithm is superior to other existing methods in speed and quality.     

An efficient kernel discriminant analysis method

Small sample size and high computational complexity are two major problems encountered when traditional kernel discriminant analysis methods are applied to high-dimensional pattern classification tasks such as face recognition. In this paper, we introduce a new kernel discriminant learning method, which is able to effectively address the two problems by using regularization and subspace decomposition techniques. Experiments performed on real face databases indicate that the proposed method outperforms, in terms of classification accuracy, existing kernel methods, such as kernel principal component analysis and kernel linear discriminant analysis, at a significantly reduced computational cost.     

一种新颖的汉字字形相似度计算方法

为了利用字形描述技术识别相似汉字, 提出了三元组递归表示的汉字字形相似度计算方法。该算法把汉字表示为汉字结构、字首部件和字尾部件三元组, 以部件为运算对象, 字型结构为运算符, 将汉字描述为前缀表达式。通过建立汉字字形相似度计算递归模型, 使计算过程被逐层分解为原子部件间的相似性比较, 从而更好地降低了计算的复杂性。然后将其用于计算汉字的最佳相似字。实验结果表明, 该方法与人的认知结果吻合度很好, 算法是有效可行的。     

基于Gabor和二值叠加CS-LBP特征的人脸表情识别

在表情识别中Gabor结合局部二值模式（LBP）的特征提取方法以及直方图统计降维虽然是较为局部化的方法,但LBP鲁棒性较差,识别精度不高,而且使用直方图统计来区分表情,其计算复杂度和特征维数依旧较高。中心对称局部二值模式（CS-LBP）与LBP相比具有较好的鲁棒性,但其对表情纹理细节的描述仍不够详细。因此提出基于Gabor结合改进的CS-LBP即二值叠加中心对称局部二值模式（二值叠加CS-LBP）的特征提取方法。用Gabor提取特征,同时用两种计算方式提取两个特征值并叠加,作为最终识别的特征;并通过离散余弦变换（DCT）降维,有效降低表情的特征维数。在JAFFE表情库中实验验证了该方法能有效提高识别精度。     

基于卷积神经网络的多人行为识别方法

为了解决多人行为识别中人物角色多且难以区分、图片增加的特征维数难以表达和学习以及行为背景复杂且容易产生干扰等问题,提出了一种基于卷积神经网络的多人行为识别方法。考虑到多人行为识别的复杂性,选择较为容易的两人交互行为作为研究对象,对实验中需要的图像数据库进行了初步的收集与预处理;然后选用在特征提取中不受拍摄角度、光照强度影响的Dense-sift算法来对原始图像进行初步的特征提取。由于人体行为图片相对手写数字图片更为复杂,因此为了使该网络能够很好地 识别 人体行为,针对该网络在其输入、网络层数、滤波器核数、学习率、输出等方面进行了修改。实验结果表明,提出的方法对拳击、拥抱、接吻3类交互行为的识别是有效的。     

一类改进型基于混沌的图像置乱网络设计

基于混沌映射的图像置乱是图像加密的一种常用方法。为了增加图像置乱网络的复杂度和提高网络运行的速度，提出了一种改进型基于混沌的图像置乱网络。该网络首先对混沌映射区间进行二次划分．然后在二次划分的区间上将混沌映射的迭代值动态量化为置乱图像的坐标，从而不仅克服了有限精度下混沌轨道遍历混沌映射区间的时间复杂度高的问题，而且减少了搜寻新坐标的迭代次数。另外，还通过计算机实验，对置乱网络的遍历时间复杂度及其置乱性质进行了分析。结果表明，这种置乱网络不仅提高了算法的运行速度，而且增加了密钥参数和密钥空间，具有良好的置乱性质，可以有效地保障加密图像的安全性。     

基于词融合与跨度检测的中文嵌套命名实体识别

目前中文命名实体识别模型在识别具有嵌套结构的实体时存在误差,无法准确识别。基于跨度的方法能够找出嵌套实体,但在识别过程中经常生成不包含实体的跨度,无法明确划分跨度边界,增加模型负担。针对此问题,提出了基于词汇融合与跨度边界检测的中文嵌套命名实体识别模型。该模型使用多词融合方法达到文本特征增强的目的,在设计的注入模块中将目标语句中字符相关的多个词汇信息进行合并,之后融入到BERT中,以此获得更全面的上下文信息,提供更好的跨度表示;其次添加跨度边界检测模块,通过感知分类器预测跨度的首尾字符来划分跨度边界。在公共数据集上的实验表明,该模型可有效提升识别准确率。     

Text/shape classifier for mobile applications with handwriting input

The paper provides a practical solution to a real-time text/shape differentiation problem for online handwriting input. The proposed structure of the classification system comprises stroke grouping and stroke classification blocks. A new set of features is derived that has low computational complexity. The method achieves 98.5 % text/shape classification accuracy on a benchmark dataset. The proposed stroke grouping machine learning approach improves classification robustness in relation to different input styles. In contrast to the threshold-based techniques, this grouping adaptation enhances the overall discriminating accuracy of the text/shape recognition system by 11.3 %. The solution improves system’s response on a touch-screen device.     

基于双边缘检测的车牌识别算法

随着智能交通的不断发展,车牌识别系统已经成为其中的重要组成部分。车牌识别分为车牌定位、字符分割以及字符识别三个部分。提出了一种新型车牌识别方法。在车牌定位方面,采用双边缘检测车牌定位方法;对于字符分割则提出了寻找连通域与传统投影分割相结合的方法;在字符识别上,将分类器分为三组,同时对于易混淆的字符进行了再次分类,这种做法缩短了训练时间,提高了准确率。实验结果表明,所提出的方法具有识别率高和速度快等特点。     

基于深度哈希的文本表示学习

文本表示学习作为自然语言处理的一项重要基础性工作, 在经历了向量空间模型、词向量模型以及上下文分布式表示的一系列发展后, 其语义表示能力已经取得了较大突破, 并直接促进模型在机器阅读、文本检索等下游任务上的表现不断提升. 然而, 预训练语言模型作为当前最先进的文本表示学习方法, 在训练阶段和预测阶段的时空复杂度较高, 造成了较高的使用门槛. 为此, 本文提出了一种基于深度哈希和预训练的新的文本表示学习方法, 旨在以更低的计算量实现尽可能高的文本表示能力. 实验结果表明, 在牺牲有限性能的情况下, 本文所提出的方法可以大幅降低模型在预测阶段的计算复杂度, 在很大程度上提升了模型在预测阶段的使用效率.     

Analysis of Mixed Korean Documents Using the Branch and Bound Algorithm Based on DP Matching

This paper presents an effective automated analysis system for mixed documents consisting of handwritten texts and graphic images. In the preprocessing step, an input image is binarized, then graphic regions are separated from text parts using chain codes of connected components. In the character recognition step, we recognize two different sets of handwritten characters: Korean and alphanumeric characters. Considering the structural complexity and variations of Korean characters, we separate them based on partial recognition results of vowels and extract primitive phonemes using a branch and bound algorithm based on dynamic programming (DP) matching. Finally, to validate recognition results, a dictionary and knowledge are employed. Computer simulation with 50 test documents shows that the proposed algorithm analyzes effectively mixed documents.     

基于对抗网络的验证码识别方法

在人机智能交互中,让机器自动识别验证码是机器模拟人的一项基础技术。基于文本的验证码识别一般先对验证码图片进行预处理,然后切割,最后对字符分类识别。字符切割的准确程度直接影响最终识别结果。提出一种对抗学习方法识别文本型验证码。先训练一个Pix2pix网络对验证码图片进行预处理,然后对抗训练出一对分割和识别网络。分割网络不仅能分割粘贴字符,而且可以筛选出难以分割的验证码结果。识别网络采用上下文相关的多通道卷积网络,能有效解决分割过程中因信息丢失而无法识别的问题。实验结果表明,该方法能提高文本验证码识别的准确率。     

基于特征点界标过滤的时间序列模式匹配方法

动态时间弯曲距离能度量不等长的时间序列、且具有较高的匹配精度,因此广泛应用在时间序列模式匹配中。但其计算复杂度较高,制约了在大规模数据集上的应用。为了实现时间序列模式度量结果和计算复杂度的平衡,提出一种基于特征点界标过滤的时间序列模式匹配方法。首先,提出一种特征点界标过滤的特征提取方法,保留时间序列主要特征,压缩时间维度;然后,利用动态时间弯曲距离对特征序列进行相似性度量;最后,在应用数据集上对所提方法进行有效性验证。实验结果表明,所提方法在保证高精度的前提下,能有效降低计算复杂度。     

Teeth recognition based on multiple attempts in mobile device

Most traditional biometric approaches generally utilize a single image for personal identification. However, these approaches sometimes failed to recognize users in practical environment due to false-detected or undetected subject. Therefore, this paper proposes a novel recognition approach based on multiple frame images that are implemented in mobile devices. The aim of this paper is to improve the recognition accuracy and to reduce computational complexity through multiple attempts. Here, multiple attempts denote that multiple frame images are used in time of recognition procedure. Among sequential frame images, an adequate subject, i.e., teeth image, is chosen by subject selection module which is operated based on differential image entropy. The selected subject is then utilized as a biometric trait of traditional recognition algorithms including PCA, LDA, and EHMM. The performance evaluation of proposed method is performed using two teeth databases constructed by a mobile device. Through experimental results, we confirm that the proposed method exhibits improved recognition accuracy of about 3.6–4.8%, and offers the advantage of lower computational complexity than traditional biometric approaches.     

基于情感角色模型的文本情感分类方法

针对传统情感分类方法因情感项指向不明引发的误判和隐藏观点遗漏等问题,提出一种基于评价对象情感角色模型的文本情感分类方法.该方法首先识别文本中的潜在评价对象,通过局部语义分析对潜在评价对象所在语句进行情感标注,确定潜在评价对象所在语句的正负极性,并定义其情感角色;然后,改进特征权值计算方法,将情感角色对应的倾向值融入模型特征空间中;最后,通过特征聚合对特征空间实现模型降维.实验结果表明,所提方法与提取强主观性情感项作为特征的情感分类方法相比,分类准确率约提高3.2%,可有效改善文本情感分类效果.     

基于CNN与有限状态自动机的手写体大写金额识别

手写票据识别是模式识别中的研究难点之一,手写体风格多样、票据背景复杂等原因导致手写票据识别的准确率不高。大写金额作为票据中最重要的部分,对其进行准确识别是手写票据自动识别的关键。对基于分割的手写体大写金额识别及处理问题进行研究,提出一种基于卷积神经网络（CNN）与有限状态自动机的手写体大写金额识别方法。在利用过分割和组合过分割项得到单字符后使用CNN对其进行识别。通过对字符进行分类、定义各类字符之间的逻辑关系构造用于语法检查的有限状态自动机,通过语法自动机在识别结果中选择符合语法规则的字符串,并在路径搜索中利用语法自动机优化搜索性能。在此基础上,运用语法自动机对模糊字符进行预测,以纠正CNN的识别错误。实验结果表明,该方法在对大写金额单字符和文本行进行识别时准确率分别高达98.2%与96.6%。     

融合Hu矩与BoF-SURF支持向量机的手势识别

基于尺度不变特征变换的特征包（BoF-SIFT）支持向量机的分类方法具有较好的手势识别效果, 但是计算复杂度高、实时性较差。为此, 提出了融合Hu矩与基于快速鲁棒特征的特征包（BoF-SURF）支持向量机（SVM）的手势识别方法。特征包模型中用快速鲁棒性特征（SURF）算法替换尺度不变特征变换（SIFT）算法提取特征, 提高了实时性, 并引入Hu矩描述手势全局特征, 进一步提高识别率。实验结果表明, 算法无论是实时性还是识别率都要高于BoF-SIFT支持向量机方法。     

基于MFCCs滤波的电话语音识别的通道补偿方法

本文提出一种基于ＭＦＣＣｓ滤波的通道补偿方法ＲＭＦＣＣ。它具有性能良好和运算简单的优点，在不失精度的前题下减少了计算代价。ＲＭＦＣＣ的性能也优于ＣＭＳ和二级ＣＭＳ。通过讨论发现许多抑制通道噪声的方法从本质上说都是采用滤波的方法，我们也证实了抑制非常低的调制频率是进行顽健的电话语音识别的有效途径。     

Spatial and temporal pyramid-based real-time gesture recognition

This paper proposes a novel method for real-time gesture recognition. Aiming at improving the effectiveness and accuracy of HGR, spatial pyramid is applied to linguistically segment gesture sequence into linguistic units and a temporal pyramid is proposed to get a time-related histogram for each single gesture. Those two pyramids can help to extract more comprehensive information of human gestures from RGB and depth video. A two-layered HGR is further exploited to further reduce the computation complexity. The proposed method obtains high accuracy and low computation complexity performance on the ChaLearn Gesture Dataset, comprising more than 50, 000 gesture sequences recorded.     

基于聚类离散化和变精度邻域熵的属性约简

针对传感网采集数据的不完备性,利用数据本身特点,通过定义类簇指标,提出基于改进K-means聚类算法的数据离散化方法,以减小噪声、孤立点和不完备数据集对决策识别结果产生的影响;然后,通过引入互信息熵的属性重要度度量和变精度修正系数,提出基于互信息熵的变精度邻域粗糙集属性约简启发式算法,整合变精度和邻域粗糙集的优势,在减小约简算法计算复杂度的同时提高决策系统识别精度.仿真结果表明了算法在提高决策系统识别精度和降低其计算复杂度方面的有效性,模拟环境测试进一步验证了其工程适用性.