基于两方向动态时间规整的无分割手写汉字检测

中文文本布局复杂,汉字种类多,书写随意性大,因而手写汉字检测是一个很有挑战的问题。本文提出了一种无分割的手写中文文档字符检测的方法。该方法用SIFT定位文本中候选关键点,然后基于关键点位置和待查询汉字大小来确定候选字符的位置,最后用两个方向动态时间规整(Dynamic Time Warping, DTW)算法来筛选候选字符。实验结果表明,该方法能够在无需将文本分割为字符的情况下准确找到待查询的汉字,并且优于传统的基于DTW字符检测方法。     

A bag-of-regions representation for video classification

A bag-of-regions (BoR) representation of a video sequence is a spatio-temporal tessellation for use in high-level applications such as video classifications and action recognitions. We obtain a BoR representation of a video sequence by extracting regions that exist in the majority of its frames and largely correspond to a single object. First, the significant regions are obtained using unsupervised frame segmentation based on the JSEG method. A tracking algorithm for splitting and merging the regions is then used to generate a relational graph of all regions in the segmented sequence. Finally, we perform a connectivity analysis on this graph to select the most significant regions, which are then used to create a high-level representation of the video sequence. We evaluated our representation using a SVM classifier for the video classification and achieved about 85 % average precision using the UCF50 dataset.     

A segmentation-free word spotting method for historical printed documents

In this paper, a two-step segmentation-free word spotting method for historical printed documents is presented. The first step involves a minimum distance matching between a query keyword image and a document page image using keypoint correspondences. In the second step of the method, the matched keypoints on the document image serve as indicators for creating candidate image areas. The query keyword image is matched against the candidate image areas in order to properly estimate the bounding boxes of the detected word instances. The method is evaluated using two datasets of different languages and is compared against segmentation-free state-of-the-art methods. The experimental results show that the proposed method outperforms significantly the competitive approaches.     

Scalable and fast SVM regression using modern hardware

Support Vector Machine (SVM) regression is an important technique in data mining. The SVM training is expensive and its cost is dominated by: (i) the kernel value computation, and (ii) a search operation which finds extreme training data points for adjusting the regression function in every training iteration. Existing training algorithms for SVM regression are not scalable to large datasets because: (i) each training iteration repeatedly performs expensive kernel value computations, which is inefficient and requires holding the whole training dataset in memory; (ii) the search operation used in each training iteration considers the whole search space which is very expensive. In this article, we significantly improve the scalability and efficiency of SVM regression by exploiting the high performance of Graphics Processing Units (GPUs) and solid state drives (SSDs). Our key ideas are as follows. (i) To reduce the cost of repeated kernel value computations and avoid holding the whole training dataset in the GPU memory, we precompute all the kernel values and store them in the CPU memory extended by the SSD; together with an efficient strategy to read the precomputed kernel values, reusing precomputed kernel values with an efficient retrieval is much faster than computing them on-the-fly. This also alleviates the restriction that the training dataset has to fit into the GPU memory, and hence makes our algorithm scalable to large datasets, especially for large datasets with very high dimensionality. (ii) To enhance the performance of the frequently used search operation, we design an algorithm that minimizes the search space and the number of accesses to the GPU global memory; this optimized search algorithm also avoids branch divergence (one of the causes for poor performance) among GPU threads to achieve high utilization of the GPU resources. Our proposed techniques together form a scalable solution to the SVM regression which we call SIGMA. Our extensive experimental results show that SIGMA is highly efficient and can handle very large datasets which the state-of-the-art GPU-based algorithm cannot handle. On the datasets of size that the state-of-the-art GPU-based algorithm can handle, SIGMA consistently outperforms the state-of-the-art GPU-based algorithm by an order of magnitude and achieves up to 86 times speedup.     

Effective and efficient classification on a search-engine model

Traditional document classification frameworks, which apply the learned classifier to each document in a corpus one by one, are infeasible for extremely large document corpora, like the Web or large corporate intranets. We consider the classification problem on a corpus that has been processed primarily for the purpose of searching, and thus our access to documents is solely through the inverted index of a large scale search engine. Our main goal is to build the “best” short query that characterizes a document class using operators normally available within search engines. We show that surprisingly good classification accuracy can be achieved on average over multiple classes by queries with as few as 10 terms. As part of our study, we enhance some of the feature-selection techniques that are found in the literature by forcing the inclusion of terms that are negatively correlated with the target class and by making use of term correlations; we show that both of those techniques can offer significant advantages. Moreover, we show that optimizing the efficiency of query execution by careful selection of terms can further reduce the query costs. More precisely, we show that on our set-up the best 10-term query can achieve 93% of the accuracy of the best SVM classifier (14,000 terms), and if we are willing to tolerate a reduction to 89% of the best SVM, we can build a 10-term query that can be executed more than twice as fast as the best 10-term query.     

基于HOG和E-SVM的服装图像联合分割算法

针对目前服装图像分割准确率低的问题,提出一种基于HOG特征和E-SVM分类器的服装图像联合分割算法。该算法具体可分为三个迭代的步骤：超像素组合、E-SVM分类器训练、分割传播,并用到辅助数据集。将用户输入的图像结合辅助服装集进行超像素分割,并利用分割传播方法将超像素组合成多个区域。利用分割效果积极的区域的HOG信息训练E-SVM分类器。通过E-SVM分类器以及分割传播方法将输入的图像中的服装分割出来。实验结果表明,该方法能够高准确率地分割出服装图像。     

Efficient query expansion with auxiliary data structures

Query expansion is a well-known method for improving average effectiveness in information retrieval. The most effective query expansion methods rely on retrieving documents which are used as a source of expansion terms. Retrieving those documents is costly. We examine the bottlenecks of a conventional approach and investigate alternative methods aimed at reducing query evaluation time. We propose a new method that draws candidate terms from brief document summaries that are held in memory for each document. While approximately maintaining the effectiveness of the conventional approach, this method significantly reduces the time required for query expansion by a factor of 5–10.     

基于HOG和SVM的双眼虹膜图像的人眼定位算法

针对近红外光下现有的人眼定位算法普遍存在准确性不高、泛化能力不佳等问题，提出了一种基于方向梯度直方图（HOG）和支持向量机（SVM）相结合的双眼虹膜图像的人眼定位算法。利用HOG提取虹膜图像的人眼特征，并结合SVM分类器对HOG特征进行训练从而实现人眼的精确定位。为了减少漏检和误检，进一步提高定位准确率，又提出了多级级联SVM分类器算法；另外针对近红外光线下虹膜图像独特的灰度分布特点，设计了一种图像预处理方法，能够显著提高人眼定位速度。在MIR2016和CASIA-IRIS-Distance数据集上的实验结果表明，基于HOG和SVM的双眼虹膜图像的人眼定位算法具有高准确率、强泛化能力和高实时性。     

A Fuzzy-Related Thesaurus for Query Refinement

Query refinement is essential for information retrieval. In this study, a fuzzy-related thesaurus based query refinement mechanism is proposed. This thesaurus can be dynamically generated during the retrieval process for a document collection that is classified by an unsupervised neural network, the self-organising map. In contrast with general relational thesaurus, the fuzzy-related thesaurus is more effective and efficient. The relationships between the terms are based on the classification of a document collection, and thus, the generated thesaurus naturally has more power to enhance retrieval quality. The recognition of the relationships can be done automatically without human involvement, which significantly reduces the cost associated with the construction of the thesaurus. An evaluation on the query refinement mechanism based on the fuzzy-related thesaurus has conducted and the preliminary result is promising. A significant improvement on retrieval performance was observed when a fuzzy-related thesaurus was used for query refinement for a software document collection.     

融入事件实体知识的汉越跨语言新闻事件检索

现有汉越跨语言新闻事件检索方法较少使用新闻领域内的事件实体知识,在候选文档中存在多个事件的情况下,与查询句无关的事件会干扰查询句与候选文档间的匹配精度,影响检索性能。提出一种融入事件实体知识的汉越跨语言新闻事件检索模型。通过查询翻译方法将汉语事件查询句翻译为越南语事件查询句,把跨语言新闻事件检索问题转化为单语新闻事件检索问题。考虑到查询句中只有单个事件,候选文档中多个事件共存会影响查询句和文档的精准匹配,利用事件触发词划分候选文档事件范围,减小文档中与查询无关事件的干扰。在此基础上,利用知识图谱和事件触发词得到事件实体丰富的知识表示,通过查询句与文档事件范围间的交互,提取到事件实体知识表示与词以及事件实体知识表示之间的排序特征。在汉越双语新闻数据集上的实验结果表明,与BM25、Conv-KNRM、ATER等基线模型相比,该模型能够取得较好的跨语言新闻事件检索效果,NDCG和MAP指标最高可提升0.712 2和0.587 2。     

HIRMA: Hypertextual information retrieval system managed by ARIOSTO

HIRMA results in an integrated environment to query any full-text document base system by natural language sentences, obtaining a document set relevant to the query. Moreover it supports hypertextual navigation into the document base. The system uses content based document representation and retrieval methods.

In this paper the representation framework as well as the retrieval and navigation algorithms used by HIRMA are described. Coverage and portability throughout application domains are supported by the lexical acquisition system ARIOSTO that provides the suitable lexical knowledge and processing methods to extract from raw text the semantic representation of documents content.     

Fast moving pedestrian detection based on motion segmentation and new motion features

The detection of moving pedestrians is of major importance for intelligent vehicles, since information about such persons and their tracks should be incorporated into reliable collision avoidance algorithms. In this paper, we propose a new approach to detect moving pedestrians aided by motion analysis. Our main contribution is to use motion information in two ways: on the one hand we localize blobs of moving objects for regions of interest (ROIs) selection by segmentation of an optical flow field in a pre-processing step, so as to significantly reduce the number of detection windows needed to be evaluated by a subsequent people classifier, resulting in a fast method suitable for real-time systems. On the other hand we designed a novel kind of features called Motion Self Difference (MSD) features as a complement to single image appearance features, e. g. Histograms of Oriented Gradients (HOG), to improve distinctness and thus classifier performance. Furthermore, we integrate our novel features in a two-layer classification scheme combining a HOG+Support Vector Machines (SVM) and a MSD+SVM detector. Experimental results on the Daimler mono moving pedestrian detection benchmark show that our approach obtains a log-average miss rate of 36 % in the FPPI range [10^?2,10⁰], which is a clear improvement with respect to the naive HOG+SVM approach and better than several other state-of-the-art detectors. Moreover, our approach also reduces runtime per frame by an order of magnitude.