The goal of image retargeting is tochange the resolution and aspect ratioof an image tofititintodifferent display devices. We achieve it by representing an image as a triangular mesh in 2D space and viewing each trian...
详细信息
The goal of image retargeting is tochange the resolution and aspect ratioof an image tofititintodifferent display devices. We achieve it by representing an image as a triangular mesh in 2D space and viewing each triangle edge as a spring. Through deforming the spring system, we can implicitly retarget the image toits new size. Tobuild the spring system,we firstly run a saliency detection algorithm togenerate each pixel's saliency. Then we get the triangular mesh after runing a Delaunay triangulation on mesh points distributed over the image according tosaliency differences. Our deformation of the spring system involves solving linear equations only once, which is more efficient than existing warp-based image retargeing methods that use iterative solver. Finally, the original image is mapped ontothe deformed spring system toget the retargeted image.
As a direction in the development of computer vision, fog visibility detection is very important for traffic *** at the visibility detection problem appearing in the highspeed road traffic, this paper proposes a fog l...
详细信息
As a direction in the development of computer vision, fog visibility detection is very important for traffic *** at the visibility detection problem appearing in the highspeed road traffic, this paper proposes a fog level detection method based on image HSV color histogram. First, convert the background image color space from RGB color space toHSV color space. And then achieve the fog level detection by classifying fog weathers of different visibility level intodifferent types using the image HSV color histogram features（including H,S, and V） in various weather conditions. The experimental results show that this method can make qualitative judgments on foggy days quickly and the qualitative detection results are relatively good.
Children's interaction picture books can not only attract children's attention but alsocan affect the recognition memory effect for content of the picture books. This paper starts with the animation medium in ...
详细信息
Children's interaction picture books can not only attract children's attention but alsocan affect the recognition memory effect for content of the picture books. This paper starts with the animation medium in interaction picture books, classifies the interactive animation in existing interaction picture books, togain samples of different degree of influence on children's attention;And memory test experiment for the selected reading samples was carried out. Experiment results show that the picture book will exert differentinfluence on children's cognitive memory effect when reading it along with the change of the attributive character of animated elements and their different ways of combination. This research can provide reference for the design and production of related children's interaction picture books.
When the number of experimental points and variables in uniform design for mixture experiments is toolarge, the requirements of uniformity and calculation efficiency are hard tobe satisfied simultaneously. In this pap...
详细信息
When the number of experimental points and variables in uniform design for mixture experiments is toolarge, the requirements of uniformity and calculation efficiency are hard tobe satisfied simultaneously. In this paper based on the transformation of U-type matrix method, the uniformity is improved by cutting method, and the calculation efficiency problem is solved by genetic algorithm. Sothe uniform design for mixture experiments with good uniformity and arbitrary number of experimental points and variables is able tobe generated. Then itis applied tothe multi-objective optimization algorithm based on physical programming toimprove optimization quality and generate evenly distributed Paretofront. Finally, the effectiveness of the improved uniform design for mixture experiments in multi-objective optimization is verified by a numerical example with three objectives.
the current Web manifests the problem of information overload due tothe success of the Web 2.0 paradigm in which users can provide new contents quickly. Tohelp users find the most valuable information, a recommendatio...
详细信息
the current Web manifests the problem of information overload due tothe success of the Web 2.0 paradigm in which users can provide new contents quickly. Tohelp users find the most valuable information, a recommendation system is designed in which we use Euclidean formula tocalculate the distance and Cosine formula tocalculate the angle todistinguish between different kinds of users. Thus, similar users will receive related items. In the beginning, we will face some problems such as the cold start due toa small amount of data. We will advance some theories tosolve the problem. With the proposed method, we can improve the quality of recommendation sothat users can find the most valuable information.
Operation-oriented Document Natural Language Understanding(take ODNLU for short) is an important approach toautomatic plotting research. However, current researches have not given a feasible method toODNLU, but with s...
详细信息
Operation-oriented Document Natural Language Understanding(take ODNLU for short) is an important approach toautomatic plotting research. However, current researches have not given a feasible method toODNLU, but with some designed processes. The purpose of this paper is toachieve ODNLU on the event level. According tothe need of automatic plotting, the event model is proposed, which contains four different classic events:configuration event, constitution event, task event, and coreference event. It describes the composition of document, and the relationship between military subjects. Then, the model identification method based on Bayes Net algorithm is presented. On the basis of these analyses, the whole ODNLU process is designed, consisting of word segment, semantic role labeling, and event model analysis. The experimental results show that this ODNLU method is feasible and effective, which achieves an average precision at 89.9%.
Cognitive scientists believe that humans memorize and understand the real world through "event". A large number of narrative class texts contain various events and people can extractimportant events from the...
详细信息
Cognitive scientists believe that humans memorize and understand the real world through "event". A large number of narrative class texts contain various events and people can extractimportant events from the texts tosupport va rious eventbased information processing. In this paper, we firstly research event annotation and build the Chinese Emergency Corpus. Then we consider the event as a basic semantic unit for narrative texts and present a new event co-occurrence network text representation method. Finally, we study important events extraction based on this event co-occurrence network. The experimental results show that our important event extraction method has good performance.
In this paper we focus on personalized recommendation algorithm for coupon deals, which are very different from deals of other retailers. We first analyzed some sample deals from Groupon and found that deals under cat...
详细信息
In this paper we focus on personalized recommendation algorithm for coupon deals, which are very different from deals of other retailers. We first analyzed some sample deals from Groupon and found that deals under category dining,wellness and activities have a high probability of having the same keywords in the deal names, which may suggest a repeated buying behavior. We believe we can use keyword matching technique for recommendation for these categories. Toprove this hypothesis, we conduct experiments on a private dataset from another company doing business similar toGroupon, and our findings from the experiments show that most of the deals from those categories bought by people have keywords matched toprevious deals bought by the same user. Finally, we propose a new recommendation algorithm which conducts keyword matching on the preliminary results from traditional collaborative filtering methods, and evaluation on this algorithm will be the further research direction.
In traditional data fusion algorithms based on context awareness, time-varying application situations and context acquisition cost are not considered, which leads toinaccurate situation prediction and low applicabilit...
详细信息
In traditional data fusion algorithms based on context awareness, time-varying application situations and context acquisition cost are not considered, which leads toinaccurate situation prediction and low applicability of data fusion. In this paper, a space-based context model is introduced, in which the sensors' history from three aspects, the context attribute, the context state and the situation space are described. Then optional attributes with the maximum overall utility are chosen by Dynamic Bayesian Networks. After that, the related situation prediction is obtained through data fusion. A data fusion algorithm CFACA(Context Fusion Algorithm based on Context Awareness) proposed in this paper gives a dynamic data fusion method. In the end, the simulation of this algorithm is discussed and the results show the effectiveness of the CFACA.
Web spamming is the deliberate manipulation osearch engine indexes tomake a page get high ranking than which it deserved considering its true value. Since the evolution of web spam, a new based on machine learning alg...
详细信息
Web spamming is the deliberate manipulation osearch engine indexes tomake a page get high ranking than which it deserved considering its true value. Since the evolution of web spam, a new based on machine learning algorithm web spam detection method which has self-learning ability has emerged. Web spam detection is viewed as a binary classification learning problem. Because labeled training examples are fairly expensive toobtain which need the participation of experts in this field and labor costs, how tofully utilize a large number ounlabeled web page examples on the web is a challenge faced by web spam detection. In this paper, we present a web spam detection algorithm according toimprove tri-training. It uses a small amount of labeled examples and a large number ounlabeled examples totrain classifiers, which can reduce the cos of labeled examples and improve the learning performance. Both web page content features and link features are used in this paper.
暂无评论