Semi-supervised classification via discriminative sparse manifold regularization

In this paper, a newly semi-supervised manifold learning algorithm named Discriminative Sparse Manifold Regularization (DSMR) is proposed. In DSMR, the whole unlabeled sample set is used to reconstruct the mean vector of each class, then obtains the sparse coefficient. For each sample of labeled samples, the new dictionary is composed of samples from the same class and the samples from the unlabeled sample set according to the corresponding rows of the sparse coefficient. For each unlabeled sample, the new dictionary is composed of samples from the whole unlabeled samples and the samples from the labeled class according to the corresponding columns of the sparse coefficient. Additionally, a discriminative term is added to stabilize performance of the algorithm. Extensive experiments on the several UCI datasets and face datasets demonstrate the effectiveness of the proposed DSMR.     

基于正则化的半监督等距映射数据降维方法

针对等距映射(ISOMAP)算法无监督,不能生成显式映射函数等局限性,该文提出一种正则化的半监督等距映射(Reg-SS-ISOMAP)算法。该算法首先利用训练样本的标签样本构建K联通图(K-CG),得到近似样本间测地线距离,并作为矢量特征代替原始数据点;然后通过测地线距离计算核矩阵,用半监督正则化方法代替多维尺度分析(MDS)算法处理矢量特征;最后利用正则化回归模型构建目标函数,得到低维表示的显式映射。算法在多个数据集上进行了比较实验,结果表明,文中提出的算法降维效果稳定,识别率高,显示了算法的有效性。     

面向图像识别的测地局部典型相关分析方法

典型相关分析(CCA)是一种经典的多模态特征学习方法,能够从不同模态同时学习相关性最大的低维特征,然而难以发现隐藏在样本空间中的非线性流形结构。该文提出一种基于测地流形的多模态特征学习方法,即测地局部典型相关分析(GeoLCCA)。该方法利用测地距离构建了低维相关特征的测地散布,并进一步通过最大化模态间的相关性和最小化模态内的测地散布学习更具鉴别力的非线性相关特征。该文不仅在理论上对提出的方法进行了分析,而且在真实的图像数据集上验证了方法的有效性。     

Video semantic analysis based on structure-sensitive anisotropic manifold ranking

As a major family of semi-supervised learning (SSL), graph-based SSL has recently attracted considerable interest in the machine learning community along with application areas such as video semantic analysis. In this paper, we analyze the connections between graph-based SSL and partial differential equation- (PDE) based diffusion. From the viewpoint of PDE-based diffusion, the label propagation in normal graph-based SSL is isotropic accompanied with distance. However, according to the structural assumption, which is one of the two basic assumptions in graph-based SSL, we need to enhance the label propagation between the samples in the same structure while weakening the counterpart between the samples in different structures. Accordingly, we deduce a novel graph-based SSL framework, named structure-sensitive anisotropic manifold ranking (SSAniMR), from PDE-based anisotropic diffusion. Instead of using Euclidean distance only, SSAniMR takes local structural difference into account to make the label propagation anisotropic, which is intrinsically different from the isotropic label propagation process in general graph-based SSL methods. Experiments conducted on the TREC Video Retrieval Evaluation (TRECVID) dataset show that this approach significantly outperforms existing graph-based SSL methods and is effective for video semantic annotation.     

Intrinsic structure based feature transform for image classification

Most dimensionality reduction works construct the nearest-neighbor graph by using Euclidean distance between images; this type of distance may not reflect the intrinsic structure. Different from existing methods, we propose to use sets as input rather than single images for accurate distance calculation. The set named as neighbor circle consists of the corresponding data point and its neighbors in the same class. Then a supervised dimensionality reduction method is developed, i.e., intrinsic structure feature transform (ISFT), it captures the local structure by constructing the nearest-neighbor graph using the Log-Euclidean distance as measurements of neighbor circles. Furthermore, ISFT finds representative images for each class; it captures the global structure by using the projected samples of these representatives to maximize the between-class scatter measure. The proposed method is compared with several state-of-the-art dimensionality reduction methods on various publicly available databases. Extensive experimental results have demonstrated the effectiveness of the proposed approach.     

基于多先验约束和一致性正则的半监督图像去雾算法

针对合成雾霾图像训练的去雾模型在真实场景中去雾效果不佳、对高层视觉任务性能提升不明显等问题,该文提出一种基于多先验约束和一致性正则的半监督图像去雾算法。该方法采用编码器-解码器网络结构,同时在合成雾霾图像与真实雾霾图像上学习去雾映射,并利用多种统计先验去雾结果作为真实雾霾图像参考真值进行半监督学习,同时通过多张真实雾霾图像的随机混合进行一致性正则约束,以消除多种先验去雾结果差异以及噪声干扰,提高图像去雾结果的视觉质量。实验对比结果表明,所提算法可比现有方法获得更好的真实场景去雾结果,并且能够显著提升高层视觉任务性能。     

基于深度学习的自然场景文本检测与识别研究

自然场景文本检测与识别技术主要应用于自动驾驶、车牌识别、智能机器人等多个场景,实用价值高,研究前景广阔。然而,自然场景的背景复杂,增加了区分文本的难度,因此相比传统的用于印刷文本检测及识别的OCR 技术,自然场景的文本检测与识别难度更高。文中提出了一种基于深度学习的自然场景文本检测与识别模型,其利用图像分割技术构建自然场景文本检测模型,并阐述了模型结构及组件。此外,还利用了压缩与激励神经网络技术来构建自然场景文本识别模型。     

Multimedia annotation via semi-supervised shared-subspace feature selection

With the rapid development of social network and computer technologies, we always confront with high-dimensional multimedia data. It is time-consuming and unrealistic to organize such a large amount of data. Most existing methods are not appropriate for large-scale data due to their dependence of Laplacian matrix on training data. Normally, a given multimedia sample is usually associated with multiple labels, which are inherently correlated to each other. Although traditional methods could solve this problem by translating it into several single-label problems, they ignore the correlation among different labels. In this paper, we propose a novel semi-supervised feature selection method and apply it to the multimedia annotation. Both labeled and unlabeled samples are sufficiently utilized without the need of graph construction, and the shared information between multiple labels is simultaneously uncovered. We apply the proposed algorithm to both web page and image annotation. Experimental results demonstrate the effectiveness of our method.     

区域信息驱动的多目标进化半监督模糊聚类图像分割算法

现有的多目标进化聚类算法应用于图像分割时,往往是在图像像素层面上进行聚类,运行时间过长,而且忽略了图像区域信息使得图像分割效果不太理想。为了提高多目标进化聚类算法的分割效果和时间效率,该文将图像区域信息与部分监督信息引入多目标进化聚类,提出图像区域信息驱动的多目标进化半监督模糊聚类图像分割算法。该算法首先利用超像素策略获得图像的区域信息,然后结合部分监督信息,设计融合区域信息和监督信息的适应度函数,接着通过多目标进化策略对多个适应度函数进行优化得到最优解集。最后构造融合区域信息与监督信息的最优解评价指标,实现从最优解集中选取一个最优解。实验结果表明：与已有多目标进化聚类算法相比,该算法不但分割效果有所提升,而且运行效率得以提高。

     

基于智能图像处理技术的车型识别

文章提出了一种应用干视频车型识别系统中的图像预处理方法，该方法对采集到的车辆图像处理后可以得到连续、光滑的边缘图像，与构建的车型库相结合，可以应用干公路车辆车型的检测与识别中。实验表明，识别过程快速、准确率较高。具有一定的参考价值。     

基于流形插值数据库构建的WLAN室内定位算法

针对传统无线局域网(WLAN)室内定位系统中因参考点密集分布及逐点信号采集所带来的位置指纹数据库构建工作量繁重的问题,该文提出一种基于混合半监督流形学习和3次样条插值的数据库构建方法。该方法利用少量标记数据和大量未标记数据求解定位目标函数的最优解,同时根据高维信号强度空间与低维物理位置空间的映射关系,实现对未标记数据的位置标定。大量实验结果表明,该方法能够在保证较高定位精度的同时,显著降低位置指纹数据库的构建开销。     

Semi-supervised feature selection with exploiting shared information among multiple tasks

Given several related tasks, multi-task feature selection determines the importance of features by mining the correlations between them. There have already many efforts been made on the supervised multi-task feature selection. However, in real-world applications, it’s noticeably time-consuming and unpractical to collect sufficient labeled training data for each task. In this paper, we propose a novel feature selection algorithm, which integrates the semi-supervised learning and multi-task learning into a joint framework. Both the labeled and unlabeled samples are sufficiently utilized for each task, and the shared information between different tasks is simultaneously explored to facilitate decision making. Since the proposed objective function is non-smooth and difficult to be solved, we also design an efficient iterative algorithm to optimize it. Experimental results on different applications demonstrate the effectiveness of our algorithm.     

基于自适应投影方法的快速车牌定位

定位是车牌识别技术的关键。针对传统投影方法的不足,提出了一种基于自适应峰区检测与投影的快速车牌定位方法,首先得到候选车牌区,然后根据车牌区的特征筛选出车牌区。实验结果表明该方法能对车牌快速准确定位。     

Automatic Image Annotation Using Semi-Supervised Multi-Instance Multi-Label Learning Algorithm

Automatic image annotation has emerged as an important research topic. From the perspective of machine learning, the annotation task fits both multiinstance and multi-label learning framework due to the fact that an image is composed of multiple regions, and is associated with multiple keywords as well. In this paper, we propose a novel Semi-supervised multi-instance multi-label （SSMIML） learning framework, which aims at taking full advantage of both labeled and unlabeled data to address the annotation problem. Specifically, a reinforced diverse density algorithm is applied firstly to select the Instance prototypes （IPs） with respect to a given keyword from both positive and unlabeled bags. Then, the selected IPs are modeled using the Gaussian mixture model （GMM） in order to reflect the semantic class density distribution. Furthermore, based on the class distribution for a keyword, both positive and unlabeled bags are redefined using a novel feature mapping strategy. Thus, each bag can be represented by one fixed-length feature vector so that the manifold-ranking algorithm can be used subsequently to propagate the corresponding label from positive bags to unlabeled bags directly. Experiments on the Corel data set show that the proposed method outperforms most existing image annotation algorithms.