Time filter

Source Type

Chi J.,Shandong University of Finance and Economics | Chi J.,Shandong Provincial Key Laboratory of Digital Media Technology | Zhang C.,Shandong University of Finance and Economics
Computer-Aided Design and Applications | Year: 2011

We present a novel automated method for capturing real-time 3D facial geometry and motion. Our method takes as input real-time face point clouds, and uses a deformable mesh model to fit each point cloud to capture the varying expressions. We propose a new normal-keeping constraint in the metric for fitting the deformable mesh to the point cloud, which not only enforces the consistency of vertex normal and intra-frame vertex motion to achieve automated capture, but also works effectively when the deformable mesh is very different from the point clouds in geometry. We also propose a new angle constraint to avoid generating poor triangles on the deformable mesh. Compared with existing techniques, our method 1) avoids manual selection of feature points, so it achieves automated capture. 2) avoids unstable optical flow estimation in traditional automated techniques, so it has high robustness. 3) always maintains good mesh structure in capture. Experiment results show the efficiency of our method. © 2011 CAD Solutions, LLC.


Liu Z.,Shandong University of Finance and Economics | Liu Z.,Shandong Provincial Key Laboratory of Digital Media Technology
Advanced Science Letters | Year: 2012

Due to the rapid increase of digital cameras and mobile phone cameras, personal photo albums have grown explosively in recent years. Effectively and accurately annotating photo albums can benefit digital photo management system. However, manually annotating all photos in an album is labor-intensive and time-consuming. Hence, this paper proposes an efficient approach to annotate photo albums with GPS information instead of annotating isolated photos. In this sense, the proposed approach mainly contains two steps: (1) Local image features and GPS information are used to construct a tripartite spectral graph, and then the photos of an album are clustered through tripartite spectral graph partitioning. (2) For a photo cluster obtained from step 1, all annotations of Corel5k dataset's vocabulary are ranked according to the relevance between each annotation and the given photo cluster, furthermore, the annotations with highest ranking score are reserved as final annotations. To compute the ranking score of a given annotation, we estimate visual similarity between the photo cluster and the images in Corel5k which have been annotated by this annotation. Experiments are conducted on the real photo albums of photo sharing website and experimental results demonstrate the effectiveness of the proposed approach. © 2012 American Scientific Publishers All rights reserved.


Shen X.,Shandong University of Finance and Economics | Shen X.,Shandong Provincial Key Laboratory of Digital Media Technology | Zhang C.,Shandong University of Finance and Economics | Zhang C.,Shandong Provincial Key Laboratory of Digital Media Technology | Wang G.,Shandong University of Finance and Economics
International Journal of Digital Content Technology and its Applications | Year: 2011

The context modeling of multiscale class label is crucial to the image segmentation algorithm based on the hidden Markov tree model (HMT). In the paper, a multi-context design for the contourletdomain HMT is proposed and applied to the texture image segmentation. At the coarse scale, the context is designed to depend on the neighboring information of class labels to strengthen the boundary segmentation. At the fine scale, the class labels of the previous coarse scale are chosen to establish the context to prevent the noisy interference. Experiments show that the segmentation algorithm based on the contourlet-domain HMT with the multiple context models outperforms the wavelet HMT segmentation algorithm in misclassification and edge preservation.


Guo Q.,Shandong University of Finance and Economics | Guo Q.,Shandong Provincial Key Laboratory of Digital Media Technology | Wei Z.,Shandong University of Science and Technology
Research Journal of Applied Sciences, Engineering and Technology | Year: 2012

This study presents a component decomposition based method for fast tire defect detection, which is motivated by the fact that defective tire images mainly consist of three components: texture, background and defect. Thus the proposed method exploits three steps to separate the defect component from the defective image. At the first step, the local total variation filtering is used to extract the texture. Then the background is estimated by the vertical mean filtering. Finally, the defect is detected by thresholding the residual image. Experimental results show that the proposed method is more accurate in locating the defects. © Maxwell Scientific Organization, 2012.


Wang L.,Hebei University of Engineering | Liu Z.,Shandong University of Finance and Economics | Liu Z.,Shandong Provincial Key Laboratory of Digital Media Technology
Przeglad Elektrotechniczny | Year: 2012

This paper presents a new hybrid method for the short-term load forecasting in electric power systems based on particle swarm optimization (PSO) and relevance vector machine (RVM). In this method, we firstly develop a type of kernel as the kernel function of the RVM model, and then its parameter is optimized by the PSO, finally the established RVM forecast mode is applied to short-term load forecasting in electric power systems in a city. The simulation results show the parameter of the wavelet kernel is well optimized using the PSO, and the acquired RVM model is more sparse and can obtain higher forecast accuracy compared with the RVM model with Gaussian kernel, so the proposed method is effective for forecasting the short-term load in electric power systems.


Liu Z.,Shandong University of Finance and Economics | Liu Z.,Shandong Provincial Key Laboratory of Digital Media Technology
Communications in Computer and Information Science | Year: 2011

With the rapid development of Web2.0 technology, we have witnessed great interest and promise in social image mining as a hot research field. Discovering and summarizing knowledge from these multimedia data enables us to mine useful information from the real world. In this paper, the approaches of three kinds of information mined from social images are reviewed: geographic information, hot events of the society and information about personal photo collections. Several key theoretical and empirical contributions in the current decade related to social image mining are discussed. Based on the analysis of what has been achieved in recent years, we believe that social image mining will be paid more and more attentions in the near future. © Springer-Verlag Berlin Heidelberg 2011.


Liu Z.,Shandong University of Science and Technology | Liu Z.,Shandong University of Finance and Economics | Liu Z.,Shandong Provincial Key Laboratory of Digital Media Technology | Ma J.,Shandong University of Science and Technology
Jisuanji Yanjiu yu Fazhan/Computer Research and Development | Year: 2011

Automatic image annotation has been an active research direction due to its great importance in content-based image retrieval(CBIR). However, the results of existing image annotation methods are still far from practical. Therefore, it is of vital importance to design a high-performance approach which could refine the initial annotations. This paper presents a novel algorithm to solve image annotation refinement problem(IAR) by graph partition and image search engine. Our algorithm focuses on pruning the noisy words in candidate annotation set to enhance image annotation performance. The main idea of the proposed algorithm lies in that candidate annotations are served as graph vertices, and the relevance between two candidate annotations is used to construct the edge weight. Then, the image annotation refinement problem can be converted to the weighted graph partition problem. The edge weight is the annotation similarity weighted by two parameters. Parameter 1 is the relationship between candidate annotation and image visual features, and parameter 2 refers to the importance of candidate annotation in host Web page. Next, we compute max cut of the graph using a heuristic algorithm. After the graph is bi-partitioned, one of the two vertex sets is chosen as final annotations. Experimental results on non-Web images and Web images show that our algorithm outperforms the existing image annotation refinement techniques.


Yan H.,Shandong University of Science and Technology | Yan H.,Shandong Provincial Key Laboratory of Digital Media Technology | Sun J.,Shandong University | Zhang C.,Shandong University of Science and Technology
Proceedings of the 2012 8th International Conference on Intelligent Information Hiding and Multimedia Signal Processing, IIH-MSP 2012 | Year: 2012

Low resolution severely degrades the detail of face images, while bad illumination condition further increases the difficulty of face recognition. In this paper, we present a new differential images based face recognition approach to deal with low resolution and variable illumination problems. Firstly, we resort to down-sampling and interpolation to build differential image database. Then feature extraction is preformed to project the differential images into differential feature subspace. For recognizing input low-resolution (LR) face images that suffer from illumination variation, normalization are employed to compensate the illumination of LR images and hallucination method is to infer the corresponding project coefficients in the differential feature subspace. Finally, similarity measure is used for face discrimination. Different from Eigenspace-based face recognition techniques, the utilization of differential images instead of original images ensures the distinction between face images of different subjects more effectively. Experiments demonstrate that the proposed approach can improve recognition rate of LR face under variable illumination significantly. © 2012 IEEE.


Chi J.,Shandong University of Finance and Economics | Chi J.,Shandong University | Chi J.,Shandong Provincial Key Laboratory of Digital Media Technology | Tu C.,Shandong University | And 2 more authors.
Visual Computer | Year: 2014

We propose a novel algorithm for the high-resolution modeling of dynamic 3D facial expressions from a sequence of unstructured face point clouds captured at video rate. The algorithm can reconstruct not only the global facial deformations caused by muscular movements, but also the expressional details generated by local skin deformations. Our algorithm consists of two parts: Extraction of expressional details and Reconstruction of expressions. In the extraction part, we extract the subtle expressional details such as wrinkles and folds from each point cloud with Laplacian smooth operator. In the reconstruction part, we use a multi-scale deformable mesh model to match each point cloud to reconstruct time-varying expressions. In each matching, we first use the low-scale mesh to match the global deformations of point cloud obtained after filtering out the expressional details, and then use the high-scale mesh to match the extracted expressional details. Comparing to many existing non-rigid ICP-based algorithms that match directly the mesh model to the entire point cloud, our algorithm overcomes the probable large errors occurred where the local sharp deformations are matched since it extracts the expressional details for separate matching, therefore, our algorithm can produce a high-resolution dynamic model reflecting time-varying expressions. Additionally, utilization of multi-scale mesh model makes our algorithm achieve high speed because it decreases iterative optimizations in matching. Experiments demonstrate the efficiency of our algorithm. © 2014 Springer-Verlag Berlin Heidelberg.


Zhao S.L.,Shandong University of Finance and Economics | Zhao S.L.,Shandong Provincial Key Laboratory of Digital Media Technology | Ji X.H.,Shandong University of Finance and Economics | Ji X.H.,Shandong Provincial Key Laboratory of Digital Media Technology
Advanced Materials Research | Year: 2014

Recently the Structural Similarity (SSIM) is proposed, and attracts a lot of attentions for its good performance and simple calculation. By deeply studying the SSIM, we find it fails to measure the badly blurred images. Based on this, we develop an improved objective quality assessment method which is based on Discrete Fourier Transform representation (called as MDFT). Experiment results show the proposed method is more consistent with HVS than SSIM especially for blurred images and fading images.

Loading Shandong Provincial Key Laboratory of Digital Media Technology collaborators
Loading Shandong Provincial Key Laboratory of Digital Media Technology collaborators