Xueqing Li

Leveraging stereopsis for saliency analysis

Stereopsis provides an additional depth cue and plays an important role in the human vision system. This paper explores stereopsis for saliency analysis and presents two approaches to stereo saliency detection from stereoscopic images. The first approach computes stereo saliency based on the global disparity contrast in the input image. The second approach leverages domain knowledge in stereoscopic photography. A good stereoscopic image takes care of its disparity distribution to avoid 3D fatigue. Particularly, salient content tends to be positioned in the stereoscopic comfort zone to alleviate the vergence-accommodation conflict. Accordingly, our method computes stereo saliency of an image region based on the distance between its perceived location and the comfort zone. Moreover, we consider objects popping out from the screen salient as these objects tend to catch a viewers attention. We build a stereo saliency analysis benchmark dataset that contains 1000 stereoscopic images with salient object masks. Our experiments on this dataset show that stereo saliency provides a useful complement to existing visual saliency analysis and our method can successfully detect salient content from images that are difficult for monocular saliency analysis methods.

Warp propagation for video resizing

This paper presents a video resizing approach that provides both efficiency and temporal coherence. Prior approaches either sacrifice temporal coherence (resulting in jitter), or require expensive spatio-temporal optimization. By assessing the requirements for video resizing we observe a fundamental tradeoff between temporal coherence in the background and shape preservation for the moving objects. Understanding this tradeoff enables us to devise a novel approach that is efficient, because it warps each frame independently, yet can avoid introducing jitter. Like previous approaches, our method warps frames so that the background are distorted similarly to prior frames while avoiding distortion of the moving objects. However, our approach introduces a motion history map that propagates information about the moving objects between frames, allowing for graceful tradeoffs between temporal coherence in the background and shape preservation for the moving objects. The approach can handle scenes with significant camera and object motion and avoid jitter, yet warp each frame sequentially for efficiency. Experiments with a variety of videos demonstrate that our approach can efficiently produce high-quality video resizing results.

Image resizing via non-homogeneous warping

Image resizing aims to adapt images to displays with different sizes and aspect ratios. In this paper, we provide a new image resizing approach for efficiently determining the non-homogeneous warp that better preserves the global image configuration and concentrates the distortion in regions of the image where they are least-likely to be noticed. Considering the different properties of large displays and small displays, we design different strategies for upsizing and downsizing. We define a variety of quadratic metrics to measure image distortion. We introduce a patch-linking scheme that can better preserve the global image configuration. We formulate image resizing as a quadratic minimization problem, which can be efficiently solved. We experiment with our method on a variety category of images and compare our results to the state of the art.

Using low-discrepancy sequences and the Crofton formula to compute surface areas of geometric models

The surface area of a geometric model, like its volume, is an important integral property that needs to be evaluated frequently and accurately in practice. In this paper, we present a new quasi-Monte Carlo method using low-discrepancy sequences for computing the surface area of a 3D object. We show that the new method is more efficient than a Monte Carlo method using pseudo-random numbers. This method is based on the Cauchy ‐ Crofton formula from integral geometry, and it computes the surface area of a 3D body B by counting the number of intersection points between the boundary surface of B and a set of straight lines in E 3 . Low discrepancy sequences are used to generate the set of lines in E 3 to reduce the estimation errors that would be caused by using statistically uniformly distributed lines. We study and compare two different methods for generating 3D random lines, and demonstrate their validity theoretically and experimentally. Experiments suggest that the new quasi-Monte Carlo method is also more efficient than the conventional approach based on surface tessellation. q 2002 Elsevier Science Ltd. All rights reserved.

A fuzzy logic based speech extraction approach for e-Learning content production

This paper proposed a fuzzy logic based approach to label speech segments from media files for content production in e-Learning systems. A fuzzy inference system (FIS) was introduced to fuzzify the extracted features extracted from different dimensions. apply predefined rules, and generate an output that represents the degree a clip belonging to speech. We show that our methodology improves the accuracy of the speech detection, and enhances the efficiency and robustness of the voice activity detection (VAD), comparing with the single feature based VAD algorithms.

Research and implementation of virtual assembly training system

With the development of human-computer interaction and virtual reality technology, virtual assembly training system shows its advantages in the field of training and education of mechanical assembly and maintenance. Based on the research of 3D interaction, virtual assembly and animation script control, this paper proposed a novel approach for designing and implementing a virtual assembly training system, including the script definition of assembly process, the playback of animation script, and the simulations of exercises and examinations. The paper described in details about the creation process of the assembly training resources, assembly study and operation exercises. Through this system, students can master the assembly process and operate the machinery in the virtual environment according to the animation and tips. The system breaks the restrictions of training equipments, training time and training place. It is convenient in large-scale and long term training, and promotes the development of mechanical assembly training.

A Constraint-based User Interface Design Method for Mobile Computing Devices

The rapid development of mobile devices and diversity of platforms require the adaptability and portability of the software. In order to automatically generate software interface on all kinds of mobile computing devices, the problem of dividing interface into screens automatically must be solved at first. This paper makes use of a constraint-based description method to describe the interface of mobile devices. In this way, the platform-independence of interface information is achieved. Taking advantage of the characteristics of interface components tree, this paper solves the problem of interface division. We discuss two division methods, depth-first principle and breadth-first principle and their advantages and disadvantages. Also the process of generating interfaces is given in this paper

Interpolation-tuned salient region detection

Image salient region detection is very useful in many multimedia applications, such as image retrieval, adaptive content delivery, and adaptive compression. Most existing methods are based on center-surround differences, usually detecting the differences around region boundaries, so these methods emphasize the high-contrast edges instead of detecting the regions. Different from traditional methods, our method smoothly propagates salient region information to the whole image by performing color interpolation and produces such saliency maps that uniformly highlight the whole salient regions and have high contrast between salient regions and backgrounds. We compare our method with five state-of-the-art salient region detection methods on a large public data set. The proposed method not only generates visually reasonable results but also achieves higher precision and better recall rates.

Design of database architecture in the saas-based multi-tenant educational information system

In traditional delivery mode of educational information systems, applications and databases are established for each college in universities independently. It not only causes resources to be wasted, but also brings about a series of problems such as data inconsistency, data redundancy, etc. In order to solve the problems mentioned above perfectly, educational information system begins to be constructed based on SaaS (Software as a service). In this paper, a schema of distributed database architecture in the SaaS-based multi-tenant educational information systems is proposed and a distributed database access process module is implemented based on J2EE technology. Through parsing and adjusting database access language semantics in the system architecture layer, the process module redirects the database access target of every tenant. In this way, the process module parses the query statement and processes the query results with the whole process being transparent to tenants. In addition, framework of application system with process module integrated into is also given in this paper.

The research and application of a SaaS-based teaching resources management platform

In the traditional teaching process management, the teaching resources management is a weak link for a long time. In this paper, an efficient solution based on SaaS (Software as a Service) idea to solve the resource publishing and fetching problem is proposed. Our solution efficiently improves the quality of teaching process. Firstly, a SaaS based business logical architecture is designed to support single-instance and multi-tenant running. In this architecture, we propose CTML (customize template markup language) to realize personalize view and utilize the authentication service to control the access of resources. Secondly, the offline publishing framework is proposed to efficiently improve the resources management. The platform proposed in this paper can improve the validity, flexibility and scalability of teaching resources management. In addition, this platform can be easily integrated to traditional education information system according to the unified authentication, and promote the further development of education informatization.