Yin Baocai

A framework and QoS based web services discovery

Current web services standards lack the support of formal semantic descriptions of a services nonfunctional characters, i.e. its Quality Of Service. When a large number of functionally-equivalent services have been discovered, it is difficult for users to choose which one to be invoked. Thus, QoS becomes a very critical factor in selecting among these candidates services to best meet users needs. In this paper, we design and build a general QoS ontology to support web services nonfunctional features. This ontology can solve the interoperability of QoS description. We also give a semantic web services discovery framework based on QoS ontology. The framework supports the automatic discovery of web services and it can improve the efficiency for users to find the best services.

Face pose estimation with a knowledge-based model

In this paper, an approach for 3D head pose estimation from a monocular sequence is proposed. At the beginning, we employ modified block matching to track face features, including five points (two at inner eyebrow tips, two at eyes and the fifth is nose tip). A matching criterion is designed the effect of matching is improved. To estimate the head pose accurately and simply, we present an algorithm based on the geometry information of the individual face and projective model without the need of any 3D face model and any special marks on users face. Any people with different facial characters can be adapted to this algorithm. At the end of this paper, some experimental results and conclusion are given.

Adaptive video coding in loop filter based on content

In order to improve the quality and efficiency of video coding and decoding, it is of great significance to select a suitable loop filter scheme. In this paper, an adaptive scheme based on content is promoted, with several elements taken into consideration, which includes the correlation between two macroblocks, the average value of motion vector in macroblock, the different level of QP value and the difference between two pixels values. All of these elements are colligated based on the statistical result of the video test and taken as the judgment criterion for strength of loop filter. The experimental results indicate that this method is good for improving the efficiency of video coding.

Real-Time Crowd Rendering and Interactions on GPU

The simulation of large crowds of characters is important in many fields of virtual reality, as they can increase the credibility of the virtual environments. Rendering large crowd of characters requires a great mount of computational power. To increase the efficiency for this render, we propose a GPU-based crowd rendering method. We present a novel hybrid rendering method using geometric representation for characters with more detail and dynamic impostors for characters with low detail. We also present a new framework to perform interaction task on GPU. The results validate that the presented method can get interactive frame rate while rendering large crowd of characters.

Spatial prediction based intra-coding [video coding]

According to the H.264 video coding standard, spatial prediction is used for intra block coding. The luma prediction may be based on a 4/spl times/4 block, for which there are nine prediction modes, or a 16/spl times/16 macroblock., for which there are four prediction modes. For chroma prediction, there are also four prediction modes. In this paper, a new method is proposed for improving the intra prediction algorithm, the first step is to apply direction to the results of the DC prediction and the second step is to use simplified modes to reduce computational complexity.

Compressive sampling and sparse reconstruction of images/videos

Vision sensors usually do not account for the physical process of imaging and they acquire image/video samples at the Nyquist rate. The Nyquist rate is significantly higher than the effective dimensions of an image/video, and consequently compression is essential for the image/video prior to storage or transmission. The emerging Compressive Sensing (CS) theory states that a signal can be perfectly reconstructed, or can be robustly approximated in the presence of noise, using a few random measurements, provided that it is sparse in some linear transform domain. CS is the theoretical foundation for capturing a signal with effective information dimensions, and thus represents an unprecedented breakthrough in many fields such as sampling, processing, and recognition of image/video. We review the fundamental problems of CS for image/video including compressive sampling, sparse reconstruction models, and algorithms for the models. For compressive sampling, the construction of random and structural measurement matrices are considered separately and the performance of these two kinds of matrices is evaluated. For sparse reconstruction, models are classified as analysis-based or synthesisbased reconstruction models by the sparse representation prior, features of which are presented. The optimization models can be considered as constrained and unconstrained optimization problems. Some feasible algorithms for these two kinds of optimization problems are explained in detail and the performance of the algorithms is given. In addition, several challenges of compressive sensing technology are presented and future work is discussed.

File format for storage of scalable video

A file format for storage of scalable video is proposed in this paper. A generic model is presented to enable a codec independent description of scalable video stream. The relationships, especially the dependencies, among sub-streams in a scalable video stream are specified sufficiently and effectively in the proposed model. Complying with the presented scalable video stream model, the file format for scalable video is proposed based on ISO Base Media File Format, which is simple and flexible enough to address the demands of scalable video application as well as the non-scalable ones.

Mesh resampling alignment for 3D face morphable model

Based on a resampling method, we present a morphable model for 3D face synthesis. The morphable model can reconstruct a 3D face automatically from a facial image by the combination of aligned prototypic 3D faces. But the point-to-point alignment of 3D dense faces is a challenging problem. The optical flow algorithm is used to compute the alignment by cylinder projective facial images. But this algorithm is an approximative and unstable method. We propose a mesh resampling method to do the alignment. Based on the resampling alignment, a 3D face morphable model is constructed. And from the temporary mesh resampling data a multi-resolution model is constructed to accelerate the model matching and improve its convergence stability. The experimental results show the resampling alignment is a contributive method for the morphable model in 3D face synthesis.

Multi-source heterogeneous data fusion method and its application in object positioning and tracking

The general sensing of Internet of Things (IoT) brings magnanimous sensing data, which shows significant multi-source heterogeneous property. How to process the multi-sourceheterogeneous sensing data intelligently and efficiently is a challenging problem. Although data fusion is considered an effective approach to processing multi-modal data and extract the hiding valuable information, there are many problems to be solved for multi-source heterogeneous data fusion, especially the unstructured video multimedia information. In this paper, the multi-source heterogeneous data fusion problem is explored and a multi-level fusion method is proposed and applied in object positioning and tracking using wireless signal, video and depth data. In the proposed method, the main opponents of data fusion are deeply studied, including the processing methods of different types of data, the feature representation and different level data fusion methods. For different types of data, from their inherent characteristics, different fusion methods at different levels are adopted here to mine the correlative relations and derive the valuable information of the heterogeneous data. The proposed method is evaluated by the object tracking and positioning experiments in actual complicated scenarios. The results show that the proposed multi-source heterogeneous data fusion based method can solve the difficulties in the traditional single dada based tracking method, such as the illumination variation, occlusion and the clutter of background. Additionally, the proposed method can estimate the three-dimensional position of the tracking object with high accuracy.

Non-uniform light field compression: a geometry-based method for image processing

A geometry-based image compression method-non-uniform light field compression is proposed by combining bi-triangle-based surface light field partition and non-uniform factoring of light field matrix, through which a four orders magnitude compression ratio can be achieved. Starting from dense images captured from vantage points, the codec resample these original data and partition the resampled data over bi-triangle. Then the codec arrange these partitioned light field data into 2D matrices and approximate the matrices through factoring them into non-uniform textures. Finally the codec group these textures into tiles and further compress these textures using ordinary still image compression techniques. Novel image can be rendered in real-time through texture mapping. In this paper we illustrate the compression efficiency and rendering performance through dense images captured in real scene.