Table of Content

23 September 2014, Volume 32 Issue 5

Previous Issue    Next Issue

Articles

Saliency Based Quality Evaluation of Point Cloud Model

ZHANG Juan1,2, ZHANG Xi-min1,2, WAN Wang-gen1,2, FANG Zhi-jun1,2

2014, 32(5):  441-446.  doi:10.3969/j.issn.0255-8297.2014.05.001

Asbtract ( 1709 )   PDF (5584KB) ( 985 )  

References | Related Articles | Metrics

 To deal with the huge amount of data in point cloud models, data must be compressed before use.Effective evaluation of quality degradation of the compressed point cloud model with respect to the original is needed. This work defines saliency of a cloud point that well reflects the human visual characteristics as the angular difference between the normal vector at the point and the average of normal vectors in its neighborhood. The difference in saliency between the compressed point cloud model and the original is taken as the objective quality index, which is easy to be calculated. Experiments on point cloud models with various resolutions indicate that the proposed method can give results conform to subjective evaluation.

Sparse Representation and Reconstruction of Image Based on Compressed Sensing

FENG Xiang1,2, WAN Wang-gen1,2

2014, 32(5):  447-452.  doi:10.3969/j.issn.0255-8297.2014.05.002

Asbtract ( 2084 )   PDF (2893KB) ( 1490 )  

References | Related Articles | Metrics

Application of compressed sensing to sparse reconstruction of image is discussed. An orthogonal matching pursuit algorithm for reconstruction and Gaussian random matrix for measurement are used. We analyze and compare DCT and DWT both theoretically and experimentally. By adjusting the sub-block size and sampling rate of the experimental images, we make a comprehensive comparison of sub-block size,sampling rate and influences of the two algorithms on effectiveness and efficiency of sparse reconstruction in terms of runtime, reconstruction error and visual effects. In sparse image representation, DCT exhibits better overall performance than DWT. In order to achieve an optimal balance between reconstruction effectiveness and efficiency, a reasonable choice of sub-block size and sampling rate is required.

Head Detection Using Skin Color and Hair Features

ZHU Qiu-yu1,2, WANG Guo-wei1, CHEN Bo1, YUAN Sai1, XU Jian-zhong1

2014, 32(5):  453-457.  doi:10.3969/j.issn.0255-8297.2014.05.003

Asbtract ( 1914 )   PDF (4835KB) ( 1379 )  

References | Related Articles | Metrics

As occlusion occurs frequently in detecting dense crowd of persons, head detection based on Adaboost performs poorly. This paper presents an approach combining skin and hair detection to achieve more reliable head detection. Skin regions in an image are extracted using several color spaces including HSV,
RGB and XYZ. Hair is extracted with a MOG model of the hair color and hair edge texture. The skin and hair regions are fused to give a head region. Experimental results show that the method has a high accuracy and can provide satisfactory results.

Point Cloud Compression Based on Compressed Sensing

ZHANG Xi-min1,2,3, YU Xiao-qing1,2, WAN Wang-gen1,2, ZHANG Juan1,2

2014, 32(5):  458-462.  doi:10.3969/j.issn.0255-8297.2014.05.004

Asbtract ( 1960 )   PDF (2177KB) ( 1215 )  

References | Related Articles | Metrics

This paper applies an algorithm for compression of 3D point cloud based on a study of scanning theory of the FARO laser scanner. The raw point cloud is processed via dimension reduction. The coordinate data are sampled and compressed, and then recovered using the orthogonal matching pursuit (OMP)algorithm.The recovered data are then processed with a statistic filter to remove edge point noise. Simulation results show that the presented method can compress point clouds with good robustness.

3D Point Cloud Deformation with Moving Least Squares and Control Curves

WANG Rui1,2, XIAO Ren1,2, CHEN Jun-li1,2, YU Zong-xin1,2, WAN Wang-gen1,2

2014, 32(5):  463-468.  doi:10.3969/j.issn.0255-8297.2014.05.005

Asbtract ( 1873 )   PDF (5279KB) ( 1196 )  

References | Related Articles | Metrics

This paper proposes a deformation method of 3D point cloud model implemented by controlling curve sets. Compared with deforming a model by controlling point sets, the proposed method makes deformed 3D point cloud model shard to be distorted, and appears more realistic. A moving least squares algorithm is used, in which curves are generated with cubic spline curve fitting. As MLS algorithm is applied to the curves to ensure rigid deformation of the 3D point cloud model. The use of control curves can achieve deformation in multiple areas accurately. Experimental results show that the method can realize smooth and realistic deformation of 3D point cloud models.

Progress in Procedural City Modeling

ZHU Xiao-qiang1,2, WAN Wang-gen1,2

2014, 32(5):  469-472.  doi:10.3969/j.issn.0255-8297.2014.05.006

Asbtract ( 1762 )   PDF (6100KB) ( 1080 )  

References | Related Articles | Metrics

The city procedural modeling techniques are investigated, which makes use of the similarities of buildings, streets and so on, in which the generation rules are defined and applied to the pre-set basic model elements. Super-large scenes can be automatically created with high efficiency, but the differences between the produced models and the real ones usually exist. So such models can be used in movies, video games and urban designs. Finally, inverse procedural modeling and its significant application in city modeling are presented.

Motion Segmentation via Multi-task Robust Principal Component Analysis

WANG Xiang-yang1,2, WAN Wang-gen1,2

2014, 32(5):  473-480.  doi:10.3969/j.issn.0255-8297.2014.05.007

Asbtract ( 1720 )   PDF (6962KB) ( 1121 )  

References | Related Articles | Metrics

This paper proposes a new algorithm, multi-task robust principal component analysis (MTRPCA),to collaboratively integrate multiple visual features for motion segmentation. Given the video data described by multiple features, the motion parts are obtained by jointly decomposing multiple feature matrices into pairs of low-rank and sparse matrices. The inference process is formulated as a convex optimization problem that minimizes a constrained combination of nuclear norm and `2,1-norm. The convex optimization problem can be solved efficiently with an augmented Lagrange multiplier (ALM) method. Compared with previous methods based on individual features, the proposed method seamlessly integrates multiple features within a single inference
step, and thus produces more accurate and reliable results. Experiments on human motion data sets, Human Eva and change detection, show that the proposed MTRPCA is effective and promising.

Medical Image Denoising Using Non-subsampled Contourlet Transform and Total Variation Model

MA Xiu-li1,2, ZHOU Feng1,2, ZHOU Xiao-jun1,2

2014, 32(5):  481-485.  doi:10.3969/j.issn.0255-8297.2014.05.008

Asbtract ( 1258 )   PDF (2053KB) ( 694 )  

References | Related Articles | Metrics

The characteristics of non-subsampled Contourlet transform (NSCT) and total variation (TV)modeling are analyzed. A mixed model of NSCT and TV is applied to medical image denoising in this paper.NSCT filter-based decomposition of noisy medical images is performed. An initial denoised image is produced using a Visu shrink threshold algorithm. The final denoised image is obtained by processing the initial denoised image with the TV model. Experimental results show that the image details are well preserved by using the proposed method. Both peak signal-to-noise ratio (PSNR) and visual quality are superior to some other denoising algorithms.

Communication Engineering

Modeling Heterogeneous Cellular Networks Based on Matern-Like Point Process Model

REN Cheng-meng, ZHANG Jian-feng, WANG Cong, XIE Wei, XU You-yun

2014, 32(5):  486-492.  doi:10.3969/j.issn.0255-8297.2014.05.009

Asbtract ( 1555 )   PDF (7284KB) ( 1251 )  

References | Related Articles | Metrics

With the increasing demands of communication and data transmission quality, traditional cellular networks containing single macrocells are gradually replaced by heterogeneous cellular networks consisting of macrocells, picocells, femtocells and relays. Stochastic point process is often used in studying base station deployment and performance analysis of heterogeneous cellular networks. The Poisson point process (PPP) is a popular one. To overcome aggregation of points in the PPP model, we introduce the Matern-like point process (MLPP), and compare it with PPP in terms of coverage probability and system efficiency. Simulation results show that MLPP has better performance with less resource, and therefore can improve system efficiency.
Taking into account that, in both PPP and MLPP, users tend to access macrocells, we introduce a bias factor to balance the load and improve the network performance.

A joint Macrocell and Femtocell Resource Allocation Based on Interference Management

XU Yue, ZHU Qi

2014, 32(5):  493-498.  doi:10.3969/j.issn.0255-8297.2014.05.010

Asbtract ( 1728 )   PDF (1761KB) ( 925 )  

References | Related Articles | Metrics

This paper proposes a joint resource allocation scheme. Joint spectrum allocation of Macrocell and Femtocells is introduced based on a dynamic heuristic clustering algorithm, which effectively mitigates cross-tier interference. An optimization problem is then formed to minimize the total power consumption of
FAPs under the QoS constraint both at each Femto user and each Macro user. A closed form of optimized power solution of FAP is derived to solve an optimization problem with a sub-gradient method. Simulation results show that the system interference and outrage of Femtocells are drastically decreased with the proposed algorithm, and therefore energy efficiency of Femtocells is enhanced.

Modeling Grid-Connected Distributed Generation

CAO Xin-hui1, SUN Yao-jie1, LIN Yan-dan1, ZHANG Jun-jun2

2014, 32(5):  499-507.  doi:10.3969/j.issn.0255-8297.2014.05.011

Asbtract ( 1009 )   PDF (8014KB) ( 640 )  

References | Related Articles | Metrics

Grid-connected distributed generation will affect security, power quality of the traditional power
system. Usually, voltage source inverters (VSI) are controlled to meet the requirements of grid-connected
distributed generation. Key technologies of the grid-connected VSI system are described in this paper. The
filter is analyzed and the design methods given. Meanwhile, the phase-locked loop (PLL) using the unbalanced
grid conditions and low voltage ride-through (LVRT) technology are studied. Based on the theoretical analysis,
a simulation model for the grid-connected Voltage source inverters (VSI) system is built and validated with
MATLAB/Simulink.

Relaying Based UWB Communication in Coal Mines

ZHANG Guo-peng, LIU Peng, DING En-jie

2014, 32(5):  508-514.  doi:10.3969/j.issn.0255-8297.2014.05.012

Asbtract ( 1297 )   PDF (1370KB) ( 552 )  

References | Related Articles | Metrics

This paper applies wireless relay technology to improve bit error rate (BER) performance of ultrawideband
(UWB) system in coal mines, and studies optimal relay deployment strategies for the lowest BERs
in different coal mine environments. The multipath propagation environment in coal mines and the resultant
BER performance for single-hop UWB links are discussed. Without increasing the transmission energy, BER
performance for multi-hop relaying based UWB transmissions is analyzed. Using the multi-hop relaying, a
decoding-and-forwarding (DF) based UWB transmission scheme is proposed. Simulation results show that,
compared with traditional single-hop UWB transmissions, the proposed method can achieve 35 dB BER
performance gain in various coal mine environments without extra energy consumption.

Signal and Information Processing

Hierarchical Structure of Deep Belief Network for Phoneme Recognition

WANG Yi1,2, YANG Jun-an1,2, LIU Hui1,2, LIU Lin3, LU Gao4

2014, 32(5):  515-522.  doi:10.3969/j.issn.0255-8297.2014.05.013

Asbtract ( 1603 )   PDF (4435KB) ( 2172 )  

References | Related Articles | Metrics

To overcome the problem of poor recognition performance and being prone to be trapped in local
optima, this paper proposes a hierarchical phoneme classification method based on deep belief network (DBN).
The system consists of two parts: a bottleneck feature and a phoneme classifier, both DBN based. The two
parts are combined to form a phoneme recognition system. The system can extract low dimensional and
supervising features, and improve classification accuracy. Experiments on TIMIT corpus suggest that the
proposed system can obtain 18.5% phoneme error rate as compared with existing systems.

Transform Domain-Based Blind Channel Estimation for MIMO-OFDM

DAI Xi-ping, PENG Hua, ZHENG Yong-jun, WANG Qi-feng

2014, 32(5):  523-529.  doi:10.3969/j.issn.0255-8297.2014.05.014

Asbtract ( 1381 )   PDF (1724KB) ( 713 )  

References | Related Articles | Metrics

The deterministic subspace algorithm is sensitive to channel order over-estimation. To deal with
the problem, a robust subspace algorithm combining a transform domain method and a subspace method is
presented, which can tolerate some order error of the channel. The proposed algorithm uses a frequency domain
subspace algorithm to estimate the channel state. A transform domain algorithm is used for noise reduction and
estimation accuracy improvement. To evaluate the performance of channel estimation, a minimum constraint
Cramer-Rao bound is derived. Theoretical analysis and simulation results show that the method have good
robustness against order over-estimation of the channel.  

Robust Algorithm for Fast Warning of Lane Departure

ZHANG Bai-rui, ZHONG Qing-hua, XUE Xiu-ting

2014, 32(5):  530-536.  doi:10.3969/j.issn.0255-8297.2014.05.015

Asbtract ( 1641 )   PDF (6107KB) ( 609 )  

References | Related Articles | Metrics

This paper proposes a robust approach for fast lane departure warning to assist safe driving. With
a monocular vision technique, binarization is realized using a region counting algorithm to enhance the image.
To find lane boundaries from the binary image, use a 515 lane template instead of the traditional edge
detection method. This method can remove much noise in the original image, improve recognition accuracy
and reduce computation load. To extract lane lines from the matched result, compressive Hough transform is
used, which reduces the required memory space as compared with traditional techniques. Experimental results
show that the proposed algorithm optimizes lane recognition and works robustly in real-time. It can also be
implemented with field programmable gate array (FPGA).

Extraction of Built-Up Areas Extraction from High-Resolution Remote-Sensing Images Using Edge Density Features

CHEN Hong, TAO Chao, ZOU Zheng-rong, SHAO Lei

2014, 32(5):  537-542.  doi:10.3969/j.issn.0255-8297.2014.05.016

Asbtract ( 1585 )   PDF (11400KB) ( 560 )  

References | Related Articles | Metrics

Built-up area contains obvious edge features. We propose a method to extract edge-based built-up
area from high resolution remote sensing images. The algorithm includes three steps: smoothing the original
image with a mean shift algorithm, extracting edges with the Canny operator and fitting them as several
straight lines, and forming a spatial voting matrix based on the edge distribution and extracting the built-up
area using the Ostu’s method. Experimental results show that the proposed approach can detect built-up areas
in images with complicated background. It is highly robust and accurate.

Steganalysis of Heterogeneous Images Using k-Means Clustering

XIE Kai, ZHANG Tao, XI Ling, LI Wen-xiang, PING Xi-jian

2014, 32(5):  543-550.  doi:10.3969/j.issn.0255-8297.2014.05.017

Asbtract ( 1555 )   PDF (3882KB) ( 693 )  

References | Related Articles | Metrics

A new image steganalysis method using k-means clustering is presented. In the training phase,
the input images are classified to several classes using k-means clustering according to texture complexity.
The training process is specialized for each class separately. In the testing phase, the given test image is first
classified to the class it belongs to according to its texture complexity. It is then submitted to the corresponding
steganalysis classifier. The proposed method can reduce mismatch penalty considerably. Experimental results
demonstrate that the method can significantly enhance detection accuracy of existing steganalysis methods.