This paper proposes a foreground moving object detection method based on Gaussian Mixture Model(GMM) and codebook. It uses GMM to extract initial foreground object, learns the background in use of codebook. It associates the foreground object obtained by GMM with the object of codebook foreground, updates the parameters of Gaussian model and codebook adaptively, according to the ratio between the foreground in 2 adjacent frames, and gets the moving object. Experimental results show that the method can eliminate the shadow of the video sequence and ghosting effectively, as well as obtain the entire foreground object in real time.
Aiming at freely editing human motion posture, Bézier curve and number data are used in human motion posture editing. According to the feature of human motion, virtual body models are constructed in the use of OpenGL in VC++. Combing the motion capture technology, the simulation of virtual people are realized. Based on the normal range of human joints, the human key joint are researched in the experiment. In the use of collected human motion data, the simulation program can drive virtual body model, which is widely used in the field of medical, education, robots simulation.
For the demand of real-time on-line route planning, an algorithm is provided here for digital terrain data compression process based on sparse quadtree’s hierarchical storage. The principle is introduced for math modeling and data compression of threats with nonformula geometry shape. The data structure and storage method are analyzed for Digital Elevation Model(DEM) data decomposed by quadtree algorithm. Another data compress algorithm is designed which is to divide the terrain data by multi-resolution sparse quadtree with the preplanned route as its center. The data updating process is described for the precision improvement of terrain data during real-time on-line application. The numerical simulation proves the mentioned validation of the method.
Traditional user models can not describe the interest of the experts in Cyberspace for Workshop of Metasynthetic Engineering(CWME) well. To meet this challenge, this paper proposes a new expert interest modeling method. This method uses Non-negative Matrix Factorization(NMF) to generate the elaborate topics, then works out the interest segments of an expert step by step through analyzing the relationship between the expert's speech and topics. The model is finally built by integrating all the interest segments of the expert. Experimental results show that the proposed model can predict discussion domain accurately and is helpful to recommend personalized information for participants in CWME.
This paper introduces the special frequency and direction properties of contrast sensitivity of human vision into the calculation of Structural Similarity(SSIM), and proposes a new structural similarity quality assessment method, called Frequency Direction Sensitive Structural Similarity(FDSSS). The value of each sub-band SSIM of the wavelet-decomposed image is calculated. To reflect the direction sensitivity of CSF, SSIM values of sub-bands under the same decomposition level with different directions are weighted according to their relative energy. To reflect the frequency sensitivity of CSF, SSIM values of each decomposition level are weighted based on the frequency magnitudes of each sub-band of CSF. Simulations results in LIVE2 and TID2008 database show that the proposed method is more consistent with subjective evaluation and outperforms than several other popular image quality assessment methods.
This paper proposes a shilling attack detection model for recommender system based on memory principle. By combining biological memory principle and mathematics statistics, it detects shilling attacks through the memory cell’s characteristics. The characteristic memory database can update timely, so that costs of system are saved. Experimental result shows that the model improves the ability of detecting shilling attacks of recommender system.
For settling a problem that Binary Space Partition(BSP) tree of in-out polygon algorithm degenerates into line list, this paper presents an improved judgment algorithm of point in-out polygon. The algorithm sorts Y value per node in polygon. It travels all horizon lines by way of binary. BSP tree is always balance binary tree. The time complexity of the BSP-Tree searching is O(lbn). Experimental results show that the improved algorithm makes sure that time complexity for the query of binary space partition trees is always high efficient while time complexity for the construction of BSP trees does not increase. Meanwhile, it has good versatility.
This paper proposes an algorithm that can authenticate image and restore tampered region by embedding the authentication watermark and restore watermark. Authentication watermark is embedded in the region of green component by slant transformed. The compressed and slant transformed coefficients as restore watermark are embeded into the least significant bit of each component. If the image is suffered tamper, the last bit of each component can also extract to achieve the image of the self-recovery. Theoretical analysis and experimental results show that under the premise of ensuring transparency and good robustness, the algorithm not only can detect the tampered area but also can restore the tempered region approximation.
It is not enough that only using a dimension analysis method for non-stable signal in a stable distribution noise. A novel wavelet time-frequency analysis method of fractional lower-order a stable distribution is proposed in this paper based on conventional wavelet, which includes the idea combining traditional wavelet and spectrum analysis. Simulations show that the proposed method has more excellent performance than conventional wavelet in time-frequency analyzing of a stable distribution process.