To identify machine and human, Completely Automated Public Turing test to tell Computers and Humans Apart (CAPTCHA) is increasingly used in many web applications. The classical English and digital characters based CAPTCHAs are recognized with high accuracy. Due to the complication of Chinese characters which greatly enhance the difficulty of automatic recognition, an increasing number of Chinese web sites use Chinese Character CAPTCHAs. To recognize Chinese Character CAPTCHAs, we propose a Convolution Neural Network (CNN) based approach to learn strokes, radicals and character features of Chinese characters, and prove that our network structure is superior to LENET-5 in this task. Furthermore, we formulate the relation among accuracy, the number of training samples and iterations, which is used to estimate the performance of our approach. Firstly, this approach greatly improves the recognition accuracy of Chinese Character CAPTCHAs with distortion, rotation and background noise. Our experiments results show that this approach achieves over 95% accuracy for single Chinese character and 84% accuracy for three types of Chinese Character CAPTCHAs with four Chinese characters. Secondly, our experiment results and theoretical analysis show that the accuracy of recognition has the exponential relationship with the product of the number of training samples and iterations in the condition of enough and representative training samples. Therefore, we can estimate the training time for a certain accuracy. Finally, we certify that our approach is superior to the most famous Chinese Optical Character Recognition (OCR) software, Hanvon, in Chinese Character CAPTCHAs recognition.

With the development of social media, an increasing number of people use short videos in social media applications to express their opinions and sentiments. However, sentiment detection of short videos is a very challenging task because of the semantic gap problem and sequence based sentiment understanding problem. In this context, we propose a SentiPair Sequence based GIF video sentiment detection approach with two contributions. First, we propose a Synset Forest method to extract sentiment related semantic concepts from WordNet to build a robust SentiPair label set. This approach considers the semantic gap between label words and selects a robust label subset which is related to sentiment. Secondly, we propose a SentiPair Sequence based GIF video sentiment detection approach that learns the semantic sequence to understand the sentiment from GIF videos. Our experiment results on GSO-2016 (GIF Sentiment Ontology) data show that our approach not only outperforms four state-of-the-art classification methods but also shows better performance than the state-of-the-art middle level sentiment ontology features, Adjective Noun Pairs (ANPs).

Understanding the sentiments of users from cross media contents which contain texts and images is an important task for many social network applications. However, due to the semantic gap between cross media features and sentiments, machine learning methods need a lot of human labeled samples. Furthermore, for each kind of media content, it is necessary to constantly add a lot of new human labeled samples because of new expressions of sentiments. Fortunately, there are some emotion signals, like emoticons, which denote users’ emotions in cross media contents. In order to use these weakly labels to build a unified multi-modality sentiment learning framework, we propose an Explicit Emotion Signal (EES) based multi-modality sentiment learning approach which uses huge number of weakly labeled samples in sentiment learning. There are three advantages in our approach. Firstly, only a few human labeled samples are needed to reach the same performance which can be obtained by the traditional machine learning based sentiment prediction approaches. Secondly, this approach is flexible and can easily combine text and vision based sentiment learning through deep neural networks. Thirdly, because a lot of weakly labeled samples can be used in EES, trained model is more robust in different domain transfer. In this paper, firstly, we investigate the correlation between sentiments and emoticons and choose emoticons as the Explicit Emotion Signals in our approach; secondly, we build a two stages multi-modality sentiment learning framework based on Explicit Emotion Signals. Our experiment results show that our approach not only achieves the best performance but also only needs 3% and 43% training samples to obtain the same performance of Visual Geometry Group (VGG) model and Long Short-Term Memory (LSTM) model in images and texts, respectively.

CAPTCHAs (Completely Automated Public Turing test to tell Computers and Humans Apart) are increasingly used in many applications for machine and human identification. Compared with traditional English and digital characters based CAPTCHAs, Chinese characters contain more complicated characters which greatly enhance difficulty of automatic recognition. To solve that problem, we proposed a Convolution Neural Network (CNN) based approach. This approach greatly improves the recognition accuracy of Chinese Character CAPTCHAs with distortion, rotation and background noise. Our experiment results show that this approach achieves more than 95% accuracy for single character and 84% accuracy for three types of Chinese Character CAPTCHAs with four characters. This encouraging result indicates that deep neural network is useful in complicated structure perception of Chinese Character CAPTCHAs.

With the development of social network applications, an increasing number of public opinion analysis systems focus on virtual network space. However, many events in virtual network space connect strongly with the events in real physical space. To solve that problem, we propose a graphical location based topic analysis framework, which combines the geographical location information and Latent Dirichlet Allocation (LDA) based topic analysis to visually analysis the connection between visual network space and real physical space. Our experiments show that this framework is useful for emergent events including natural disaster and social unrest.

Processing of web text clustering data usually results in more than ten thousand features. The traditional dimensionality reduction methods and optimal feature subset selection methods increase the time complexity. To reduce the time complexity, we propose a Minimum Information Quantity Partition (MIQP) method. First, MIQP selects a useful feature subset by determining the best partition according to the diminishing trend of feature weight curve. Second, to remove the feature independent assumption and compute the semantic relation between selected features, Latent Semantic Indexing (LSI) is used to eliminate noisy data and extend the missed semantic of each sample. This approach reduces the time complexity from O(mn3) to O(mn2). The experimental results show that the performance of MIQP is close to the best clustering results of selecting top k features, and the speed of MIQP is much faster than clustering with all features in our experiment data.

Rumor is one of the important issues for social media. Previous works mainly focus on using social attribute features in rumor analysis. However, social attribute features don't indicate the purpose of a rumor which is one of the most important aspects of a rumor. To solve that problem, we focus on not only those social attribute features, but also social content features to find out what kind of features are useful for exploring the purpose of a rumor. Finally, we propose 6 kinds of features, where four of them belong to social attribute features and two of them belong to social content features. To uncover the purpose of rumors from proposed features, we choose Sina weibo, the biggest micro-blog platform in China, and crawl 11,676 rumors for analysis. The analysis results show that the diffusion purpose of rumors can be concluded from social content attributes, and proposed two layers KL divergence approach is useful in diffusion purpose words perception.

ECG is a kind of high dimensional dataset and the useful information of illness only exists in few heartbeats. To achieve a good classification performance, most existing approaches used features proposed by human experts, and there is no approach for automatic useful feature extraction. To solve that problem, we propose an ECG Codebook Model (ECGCM) which automatically builds a small number of codes to represent the high dimension ECG data. ECGCM not only greatly reduces the dimension of ECG, but also contains more meaningful semantic information for Myocardial Infarction detection. Our experiment results show that ECGCM achieves 2% and 20.5% improvement in sensitivity and specificity respectively in Myocardial Infarction detection.

The self-training based discriminative tracking methods use the classification results to update the classifier itself. However, these methods easily suffer from the drifting issue because the classification errors are accumulated during tracking. To overcome the disadvantages of self-training based tracking methods, a novel co-training tracking algorithm, termed Co-SemiBoost, is proposed based on online semi-supervised boosting. The proposed algorithm employs a new online co-training framework, where unlabeled samples are used to collaboratively train the classifiers respectively built on two feature views. Moreover, the pseudo-labels and weights of unlabeled samples are iteratively predicted by combining the decisions of a prior model and an online classifier. The proposed algorithm can effectively improve the discriminative ability of the classifier, and is robust to occlusions, illumination changes, etc. Thus the algorithm can better adapt to object appearance changes. Experimental results on several challenging video sequences show that the proposed algorithm achieves promising tracking performance.

There exist plenty uncorrelated features in the high dimensional ECG data,so,it is difficult for the classifier based on supervised learning to perform well in both sensitivity and specificity.Pre-processed by baseline wander removing,high-frequency span removing and polynomial fitting,an auto heartbeat classification model is proposed based on supervised MCA dimension reducing.The sequence ECG data is discretized;supervised MCA dimension reducing technology is employed to extract the key features;the ECG data is classified with the common classifiers.The experiment on the PTB database shows,compared with supervised learning method,this approach combining with different classifiers has a better performance on both sensitivity and specificity.

This paper presents a new latent topic multiple instance learning (LTMIL) for automated ECG classification. Due to the characteristics of multiple beats constituting an ECG and the high cost of having all the beats manually labeled, supervised machine learning techniques have achieved limited success in ECG classification. In this paper, we first discuss the rational for applying multiple instance learning (MIL) to automated ECG classification and propose a new MIL strategy called LTMIL for it which integrates the intra and inter ECG difference. It is a three hierarchical model. Firstly, we cluster all unlabeled beats into k topics using variable weighting, by which beats are reshaped to be dense and separable. An ECG and its beats are then represented as mixtures over topics. Consequently, the intra and inter ECG difference can be fully embodied in the difference between mixtures over topics. Finally, any supervised learning techniques can be applied to classification of transformed ECGs. Our experimental results on real ECG datasets from the PTB diagnostic database demonstrate that compared with existing multiple instance learning and supervised machine learning algorithms, the proposed algorithm is able to automatically classify ECG without labeling beats and improves the classification quality in terms of sensitivity and specificity.

Wavelet analysis based on image enhancement technique is only applicable to black-and-white image, and pseudo-color image processing technology cannot adequately deal with some of the details information of the image. In this paper, an enhanced approach of remote sensing image based on orthogonal wavelet analysis and pseudo-color image processing is presented. Enhanced remote sensing image has been greatly improved in both visual effects and noise characteristics. The method is simple yet flexible with less calculation. Moreover, the proposed method also has the advantages of faster computing speed and operating feasibly and so on. It has great potential in research and application of remote sensing image enhancement.

This paper presents our recent research work on a remote diagnoses system for colonic cancer in pervasive environment. In the system, the high quality of multiple protein sequence alignment for patients plays a crucial role in doctors' correct decision-making. This paper laid stress on improving the performance of multiple sequence alignment. Two improved evolutionary algorithms are proposed here. One is based on a genetic algorithm, where segment profiles are introduced to speed up convergence. The other one is an application of an improved particle swarm optimization algorithm, where the principles of information diffusion and clone selection are incorporated to prevent premature convergence. The two new algorithms are compared with the ClustalX and T-Coffee programs on several data cases from the BAHBASE benchmark alignment database. The experimental results show that they can yield better performance on data sets and suit multiple alignment of protein sequences with different length and similarity

This paper presents a novel genetic algorithm (GA) for multiple sequence alignment in protein analysis. The most significant improvement afforded by this algorithm results from its use of segment profiles to generate the diversified initial population and prevent the destruction of conserved regions by crossover and mutation operations. Segment profiles contain rich local information, thereby speeding up convergence. Secondly, it introduces the use of the norMD function in a genetic algorithm to measure multiple alignment Finally, as an approach to the premature problem, an improved progressive method is used to optimize the highest-scoring individual of each new generation. The new algorithm is compared with the ClustalX and T-Coffee programs on several data cases from the BAliBASE benchmark alignment database. The experimental results show that it can yield better performance on data sets with long sequences, regardless of similarity.

A novel PSO algorithm called InformPSO is introduced in this paper. The premature convergence problem is a deficiency of PSOs. First, we analyze the causes of premature convergence for conventional PSO. Second, the principles of information diffusion and clonal selection are incorporated into the proposed PSO algorithm to achieve a better diversity and break away from local optima. Finally, when compared with several other PSO variants, it yields better performance on optimization of unimodal and multimodal benchmark functions.

针对直放站集群系统的负载均衡问题，本文提出基于连接等级的直放站集群调度算法。该方法综合经典LVS调度算法中的服务器权重和当前连接数两个负载均衡因素形成连接等级值，进而通过等级值确定选定的服务器。本算法的特点是有效缩短了服务器的调度时间并提高负载均衡性能。实验结果表明该算法在调度时间上优于经典的LVS调度算法。

针对花卉类间相似和类内差异的问题，传统的特征融合只是简单的把多个特征融合在一起，忽略了不同特征对不同的花卉类别的影响。为解决这一问题，本文提出一种基于层次化特征融合的分类方法。该方法针对每类花卉的颜色、形状特征进行单独训练，得到特征对应于类别的得分值，能够对表现差异大的花卉进行有效的分类，同时又能够克服类内变化大的问题。实验结果表明，该方法比直接的特征融合方式分类效果更好，能够有效地分类出类间相似度比较高以及类内差异性比较大的花卉。

针对花卉图像复杂的背景问题，本文提出基于显著性检测的花卉图像前景分割方法。该方法通过为花卉图像的显著性区域训练前景背景分类器，自适应设定初始前景背景的信息分布，并结合GrabCut算法实现将花卉的主体部分从背景中分离开。相对于经典的花卉图像分割方法。本文提出的方法，不需要对每一类花进行单独的训练，是一种自适应的花卉图像分割方法。