Approximate regular expression pattern matching with concave gap penalties

Given a sequenceA of lengthM and a regular expressionR of lengthP, an approximate regular expression pattern-matching algorithm computes the score of the optimal alignment betweenA and one of the sequencesB exactly matched byR. An alignment between sequencesA=a₁a₂ ... a_M andB=b₁b₂... b_N is a list of ordered pairs, (i₁,j₁), (i₂j₂), ..., (i_t,j_tt) such that i_k < i_k+1 and j_k < j_k+1. In this case the alignmentaligns symbols a_ik and b_jk, and leaves blocks of unaligned symbols, orgaps, between them. A scoring schemeS associates costs for each aligned symbol pair and each gap. The alignment's score is the sum of the associated costs, and an optimal alignment is one of minimal score. There are a variety of schemes for scoring alignments. In a concave gap penalty scoring schemeS={, w}, a function (a, b) gives the score of each aligned pair of symbolsa andb, and aconcave function w(k) gives the score of a gap of lengthk. A function w is concave if and only if it has the property that, for allk > 1, w(k + 1) –w(k) w(k) –w(k –1). In this paper we present an O(MP(logM + log² P)) algorithm for approximate regular expression matching for an arbitrary and any concavew. This work was supported in part by the National Institute of Health under Grant RO1 LM04960.     

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。     

A Multi-scale Attention-based Facial Emotion Recognition Method Based on Deep Learning

Recently, people have been paying more and more attention to mental health, such as depression, autism, and other common mental diseases. In order to achieve a mental disease diagnosis, intelligent methods have been actively studied. However, the existing models suffer the accuracy degradation caused by the clarity and oc-clusion of human faces in practical applications. This paper, thus, proposes a multi-scale feature fusion network that obtains feature information at three scales by locating the sentiment region in the image, and integrates global feature information and local feature information. In addition, a focal cross-entropy loss function is designed to improve the network''s focus on difficult samples during training, enhance the training effect, and increase the model recognition accuracy. Experimental results on the challenging RAF_DB dataset show that the proposed model exhibits better facial expression recognition accuracy than existing techniques.     

基于改进轻量级秩扩展网络的人脸表情识别方法

人脸表情识别作为人机交互的一种重要方法,广泛应用于智能医疗、公安测谎系统、车载安全系统等领域。现有人脸表情识别方法多数存在参数量冗余、计算成本高、特征表达瓶颈等问题。提出一种基于改进轻量级秩扩展网络ReXNet的人脸表情识别方法。通过构建改进的ReXNet以提取人脸表情特征,在参数量较少的条件下解决特征表达瓶颈的问题,增强对表情局部特征的关注,获得高层次的表情特征,同时融合坐标注意力模块,将位置信息嵌入到通道注意力中,精准地定位和识别感兴趣的特征,建立位置信息与局部特征之间的长依赖关系,减少计算开销。在此基础上,将细化模块引入到改进的网络架构中,利用类别上下文信息细化分类结果,增强类间的分化效果,从而提高人脸表情识别的准确率。实验结果表明,该方法在RAF-DB和FERPlus数据集上的人脸表情识别准确率分别达到88.43%和88.8%,相比VGG16-PLD、SHCNN、ResNet+VGG等方法,具有较高的准确率和较优的鲁棒性。     

基于图注意力网络字词融合的中文命名实体识别

命名实体识别指识别文本中具有特定意义的实体,是自然语言处理诸多下游任务的重要基石。在命名实体识别任务中,协同图网络（CGN）模型通过引入多个图注意力网络获得较强的知识整合能力及较高的处理速度,但CGN模型在嵌入层中没有充分利用词边界信息,且采用的传统静态图注意力网络影响了图注意力的表达能力。在对CGN模型的图注意力网络进行改进的基础上,提出一种中文命名实体识别新模型,在嵌入层融入词语的分词信息,以生成包含词边界信息的字向量,从而充分利用词边界信息。通过在编码层使用BiLSTM模型获取文本的上下文信息,采用改进后的图注意力网络提取文本特征,并通过优化传统图注意力网络中相关系数的计算方式,增强模型的特征提取能力。最后,利用条件随机场对文本进行解码,从而实现对实体的标注。实验结果表明,该模型相比CGN模型在MSRA、OntoNotes4.0、Weibo数据集上的F1值分别提升了0.67%、3.16%、0.16%,验证了其在中文命名实体识别任务上的有效性。     

基于有向图模型的旅游领域命名实体识别

旅游领域命名实体识别是旅游知识图谱构建过程中的关键步骤,与通用领域的实体相比,旅游文本的实体具有长度长、一词多义、嵌套严重的特点,导致命名实体识别准确率低。提出一种融合词典信息的有向图神经网络（L-CGNN）模型,用于旅游领域中的命名实体识别。将预训练词向量通过卷积神经网络提取丰富的字特征,利用词典构造句子的有向图,以生成邻接矩阵并融合字词信息,通过将包含局部特征的词向量和邻接矩阵输入图神经网络（GNN）中,提取全局语义信息,并引入条件随机场（CRF）得到最优的标签序列。实验结果表明,相比Lattice LSTM、ID-CNN+CRF、CRF等模型,L-CGNN模型在旅游和简历数据集上具有较高的识别准确率,其F1值分别达到86.86%和95.02%。     

用于嵌套命名实体识别的边界强化分类模型

实体嵌套是自然语言中一种常见现象,提高嵌套命名实体识别的准确性对自然语言处理各项任务具有重要作用。针对现有嵌套命名实体识别方法在识别实体边界时不够准确、未能有效利用实体边界信息等问题,提出一种嵌套命名实体识别的边界强化分类模型。采用卷积神经网络提取邻接词的特征,通过加入多头注意力的序列标注模型获取实体中的边界特征,提高实体边界检测的准确性。在此基础上,计算实体中各词语对实体类型的贡献度,将实体关键字与实体边界词相结合来表示实体,使实体表示中包含实体关键信息和边界信息,最后进行实体类型检测。实验结果表明,通过加入多头注意力机制能够有效提升对嵌套命名实体的检测和识别性能,该模型在GENIA和GermEval 2014数据集上准确率有较好表现,并且召回率和F1值较对比模型达到最优。     

Operator context scanning to support high segmentation rates for real time license plate recognition

Introducing high definition videos and images in object recognition has provided new possibilities in the field of intelligent image processing and pattern recognition. However, due to the large amount of information that needs to be processed, the computational costs are high, making the HD systems slow. To this end, a novel algorithm applied to sliding window analysis, namely Operator Context Scanning (OCS), is proposed and tested on the license plate detection module of a License Plate Recognition (LPR) system. In the LPR system, the OCS algorithm is applied on the Sliding Concentric Windows pixel operator and has been found to improve the LPR system’s performance in terms of speed by rapidly scanning input images focusing only on regions of interest, while at the same time it does not reduce the system effectiveness. Additionally, a novel characteristic is presented, namely, the context of the image based on a sliding windows operator. This characteristic helps to quickly categorize the environmental conditions upon which the input image was taken. The algorithm is tested on a data set that includes images of various resolutions, acquired under a variety of environmental conditions.     

Personalized text snippet extraction using statistical language models

In knowledge discovery in a text database, extracting and returning a subset of information highly relevant to a user's query is a critical task. In a broader sense, this is essentially identification of certain personalized patterns that drives such applications as Web search engine construction, customized text summarization and automated question answering. A related problem of text snippet extraction has been previously studied in information retrieval. In these studies, common strategies for extracting and presenting text snippets to meet user needs either process document fragments that have been delimitated a priori or use a sliding window of a fixed size to highlight the results. In this work, we argue that text snippet extraction can be generalized if the user's intention is better utilized. It overcomes the rigidness of existing approaches by dynamically returning more flexible start-end positions of text snippets, which are also semantically more coherent. This is achieved by constructing and using statistical language models which effectively capture the commonalities between a document and the user intention. Experiments indicate that our proposed solutions provide effective personalized information extraction services.     

Novel moment invariants for improved classification performance in computer vision applications

A novel set of moment invariants based on the Krawtchouk moments are introduced in this paper. These moment invariants are computed over a finite number of image intensity slices, extracted by applying an innovative image representation scheme, the image slice representation (ISR) method. Based on this technique an image is decomposed to a several non-overlapped intensity slices, which can be considered as binary slices of certain intensity. This image representation gives the advantage to accelerate the computation of image's moments since the image can be described in a number of homogenous rectangular blocks, which permits the simplification of the computation formulas. The moments computed over the extracted slices seem to be more efficient than the corresponding moments of the same order that describe the whole image, in recognizing the pattern under processing. The proposed moment invariants are exhaustively tested in several well known computer vision datasets, regarding their rotation, scaling and translation (RST) invariant recognition performance, by resulting to remarkable outcomes.