Hock Soon Seah

Nanyang Technological University, Tumasik, Singapore

Are you Hock Soon Seah?

Claim your profile

Publications (102)46.08 Total impact

  • [Show abstract] [Hide abstract]
    ABSTRACT: 3D human motion tracking has received increasing attention in recent years due to its broad applications. Among various 3D human motion tracking methods, the particle filter is regarded as one of the most effective algorithms. However, there are still several limitations of current particle filter approaches such as low prediction accuracy and sensitivity to discontinuous motion caused by low frame rate or sudden change of human motion velocity. Targeting such problems, this paper presents a full-body human motion tracking system by proposing exemplar-based conditional particle filter (EC-PF) for monocular camera. By introducing a conditional term with respect to exemplars and image data, dynamic model is approximated and used to predict current states of particles in prediction phase. In update phase, weights of particles are refined by matching images with projected human model using a set of features.
  • [Show abstract] [Hide abstract]
    ABSTRACT: Curve extension is a useful function in CAD systems. Disk B-Spline curve has its distinct advantages in representing a 2D region. This paper presents an algorithm for extending the disk B-Spline curve with G2 continuity. A disk Bezier segment is used to construct the extending part and G2-continuity can be used to describe the smoothness at the joint disk. Fairness of the extending disk Bezier curve segment is achieved by minimizing energy objective functions for the center curve and the radius function separately. New control disks are computed by unclamping algorithm to represent the whole extended disk B-Spline curve. The experimental results demonstrate the effectiveness of our method.
    Computer-Aided Design and Applications 03/2015; DOI:10.1080/16864360.2015.1014729
  • Source
    Yaqiong Liu, Hock Soon Seah
    International Journal of Geographical Information Science 03/2015; DOI:10.1080/13658816.2015.1005094 · 1.48 Impact Factor
  • [Show abstract] [Hide abstract]
    ABSTRACT: Since the human visual system (HVS) is the ultimate appreciator of most photorealistically rendered images, rendering process can be accelerated by exploiting the properties of the HVS. According to the concept of entropy masking, the HVS is not sensitive to visual distortions in unstructured visual signals. For structured regions, pixels are highly correlated, while the similarity among pixels in unstructured regions is low. In this paper, we detect unstructured regions by extracting local patches from each pixel and its neighboring pixels, and comparing the similarity between the local patches of the center pixel and the neighboring pixels. We further exploit entropy masking in perceptual rendering, and experimental results demonstrate that the proposed method can accelerate rendering, without degrading the perceived quality of resultant images.
    Signal Processing Image Communication 02/2015; 33. DOI:10.1016/j.image.2015.02.001 · 1.15 Impact Factor
  • [Show abstract] [Hide abstract]
    ABSTRACT: Erratum to: Vis Comput DOI 10.1007/s00371-014-1037-zThe published paper contains a minor error (note the section highlighted with bold formatting) in the Abstract which we wish to make a correction on:\(\ldots \) Secondly, by considering the surrounding objects (their material reflectance from and visible surface area of the caustic objects) and light power, we compute the refinement amount of each wavelength cluster. Our accelerated \(\ldots \)Should be written as:\(\ldots \) Secondly, by considering the surrounding objects (their material reflectance and their visible surface area from the caustic objects) and light power, we compute the refinement amount of each wavelength cluster. Our accelerated \(\ldots \)
    The Visual Computer 03/2014; 31(3):365-365. DOI:10.1007/s00371-014-1051-1 · 1.07 Impact Factor
  • [Show abstract] [Hide abstract]
    ABSTRACT: Curve blending is an essential task in geometric modeling, while a ball B-spline curve (BBSC) has its advantages in representing freeform tubular objects. This paper proposes a blending algorithm for ball B-Spline curve with G2 continuity, which is used to describe the smoothness of the joint point. An original BBSC is extended smoothly to join another one, such that no additional blending curve is created and the two original curves are not changed. The shape of the extended curve is then determined by minimizing strain energy. The corresponding scalar function of the control balls is determined through applying G2-continuity conditions to the scalar function. In order to ensure the radii of the control balls are positive, we make a decision about the range of the G2-continuity parameter and then determine it by minimizing the strain energy in the affected area. The experiment results demonstrate our method for blending BBSC is effective. Moreover, some G2 blending results of the BBSC in simulating the tubular objects are given.
    Proceedings of the 2013 International Conference on Computer-Aided Design and Computer Graphics; 11/2013
  • [Show abstract] [Hide abstract]
    ABSTRACT: Curve extension is a useful function in CAD systems. Disk B-Spline curve has its distinct advantages in representing a 2D region. This paper presents an algorithm for extending the disk B-Spline curve. A disk Bezier segment is used to construct the extending part and G2-continuity can be used to describe the smoothness of the joint disk. Fairness of the extending disk Bezier segment is achieved by minimizing an energy objective function. New control disks are computed by unclamping algorithm to represent the whole extended disk B-Spline curve. The experimental results demonstrate the effectiveness of our method.
    Proceedings of the 2013 International Conference on Computer-Aided Design and Computer Graphics; 11/2013
  • Yaqiong Liu, Hock Soon Seah, Gao Cong
    [Show abstract] [Hide abstract]
    ABSTRACT: Given a set of moving clients as well as their friend relationships, a road network, and a distance threshold per friend pair, the proximity detection problem in road networks is to find each pair of friends such that the road network distance between them is within the given threshold. The problem of proximity detection is often encountered in friend-locator applications and massively multiplayer online games. Because of the limited battery power and bandwidth, it is better to develop a solution which incurs less communication cost. Hence, the main objective of this problem is to reduce the total communication cost. However, most of the existing proximity detection solutions focus on the Euclidean space but cannot be used in road network space; the solutions for road networks incur substantial communication costs. Motivated by this, we propose two types of solutions to solve the proximity detection problem in road networks. In the first type of solution, each mobile client is assigned with a mobile region of a fixed size. We design algorithms with a fixed radius for the client and server respectively, with the purpose of reducing unnecessary probing messages and update messages. Second, we present a self-tuning policy to adjust the radius of the mobile region automatically to minimize the communication cost. Experiments show that our second type of solution works efficiently and robust with a much lower communication cost with respect to various parameters. In addition, we present our server-side computational cost optimization techniques to reduce the total computational cost.
    Proceedings of the 21st ACM SIGSPATIAL International Conference on Advances in Geographic Information Systems; 11/2013
  • [Show abstract] [Hide abstract]
    ABSTRACT: Our research focuses on the problem of path planning in 3D virtual world applications. The characters we consider are heterogeneous, as they have different sizes, and can perform surface or volumetric motion. In this paper, we propose an enhanced waypoint graph, which consists of point nodes equipped with radius, as well as edges connecting those nodes. Each edge is labeled with the motion type it can support. Given a polygon soup representation of a virtual world, the proposed algorithm starts by subdividing the virtual world into regions. This enables us to process large virtual worlds. Each region is then locally voxelized, one at a time. Two kinds of waypoints are generated: local waypoints using corner detection on the voxelization, and border waypoints at the region boundary. Waypoints are then sparsely connected to form a local waypoint graph, and local graphs are finally connected via the border waypoints to create the final global enhanced waypoint graph. To plan paths between arbitrary points using this graph, the points are connected to the graph using nearest neighbor search and traversability test, then Dijkstra/A* algorithm is used to calculate the final path, taking into account the appropriate size and motion type.
    The Visual Computer 10/2013; 29(10). DOI:10.1007/s00371-013-0837-x · 1.07 Impact Factor
  • X. Xu, H. S. Seah, C. K. Quah
    [Show abstract] [Hide abstract]
    ABSTRACT: Producing traditional animation is a laborious task where the key drawings are first drawn by artists and thereafter inbetween drawings are created, whether it is by hand or computer‐assisted. Auto‐inbetweening of these 2D key drawings by computer is a non‐trivial task as 3D depths are missing. An alternate approach is to generate all the drawings by extracting lines directly from animated 3D models frame by frame, concatenating and rendering them together into an animation. However, animation quality generated using this straightforward method bears two problems. Firstly, the animation contains unsatisfactory visual artifacts such as line flickering and popping. This is especially pronounced when the lines are extracted using high‐order derivatives, such as ridges and valleys, from 3D models represented in triangle meshes. Secondly, there is a lack of temporal continuity as each drawing is generated without taking its neighboring drawings into consideration. In this paper, we propose an improved approach over the straightforward method by transferring extracted 3D line drawings of each frame into individual 3D lines and processing them along the time domain. Our objective is to minimize the visual artifacts and incorporate temporal relationship of individual lines throughout the entire animation sequence. This is achieved by creating correspondent trajectory of each line from each frame and applying global optimization on each trajectory. To realize this target, we present a fully automatic novel approach, which consists of (1) a line matching algorithm, (2) an optimizing algorithm, taking into account both the variations of numbers and lengths of 3D lines in each frame, and (3) a robust tracing method for transferring collections of line segments extracted from the 3D models into individual lines. We evaluate our approach on several animated model sequences to demonstrate its effectiveness in producing line drawing animations with temporal coherence.
    Computer Graphics Forum 10/2013; 32(7). DOI:10.1111/cgf.12236 · 1.60 Impact Factor
  • [Show abstract] [Hide abstract]
    ABSTRACT: We present a surface reconstruction via 2D strokes and a vector field on the strokes based on a two-step method. In the first step, from sparse strokes drawn by artists and a given vector field on the strokes, we propose a nonlinear vector interpolation combining total variation and H 1 regularization with a curl-free constraint for obtaining a dense vector field. In the second step, a height map is obtained by integrating the dense vector field in the first step. Jump discontinuities in surface and discontinuities of surface gradients can be well reconstructed without any surface distortion. We also provide a fast and efficient algorithm for solving the proposed functionals. Since vectors on the strokes are interpreted as a projection of surface gradients onto the plane, different types of strokes are easily devised to generate geometrically crucial structures such as ridge, valley, jump, bump, and dip on the surface. The stroke types help users to create a surface which they intuitively imagine from 2D strokes. We compare our results with conventional methods via many examples.
    Numerical Mathematics Theory Methods and Applications 02/2013; 6(1). DOI:10.4208/nmtma.2013.mssvm16 · 0.77 Impact Factor
  • [Show abstract] [Hide abstract]
    ABSTRACT: Compared to monocular pose tracking, 3D articulated body pose tracking from multiple cameras can better deal with self-occlusions and meet less ambiguities. Though considerable advances have been made, pose tracking from multiple images has not been extensively studied: very seldom existing work can produce a solution comparable to that of a marker-based system which generally can recover accurate 3D full-body motion in real-time. In this paper, we present a multi-view approach to 3D body pose tracking. We propose a pose search method by introducing a new generative sampling algorithm with a refinement step of local optimization. This multi-layer search method does not rely on strong motion priors and generalizes well to general human motions. Physical constraints are incorporated in a novel way and 3D distance transform is employed for speedup. A voxel subject-specific 3D body model is created automatically at the initial frame to fit the subject to be tracked. We design and develop the optimized parallel implementations of time-consuming algorithms on GPU (Graphics Processing Unit) using CUDA (Compute Unified Device Architecture), which significantly accelerates the pose tracking process, making our method capable of tracking full body movements with a maximum speed of 9 fps. Experiments on various 8-camera datasets and benchmark datasets (HumanEva-II) captured by 4 cameras demonstrate the robustness and accuracy of our method.
    IEEE Transactions on Multimedia 01/2013; 15(1):106-119. DOI:10.1109/TMM.2012.2225040 · 1.78 Impact Factor
  • [Show abstract] [Hide abstract]
    ABSTRACT: In this work, we propose a novel graphic saliency detection method to detect visually salient objects in images rendered from 3D geometry models. Different from existing graphic saliency detection methods, which estimate saliency based on pixel-level contrast, the proposed method detects salient objects by computing object-level contrast. Given a rendered image, the proposed method first extracts dominant colors from each object, and represents each object with a dominant color descriptor (DCD). Saliency of each object is then calculated by measuring the contrast between the DCD of the object and the DCDs of its surrounding objects. We also design a new iterative suppression operator to enhance the saliency result. Compared with existing graphic saliency detection methods, the proposed method can obtain much better performance in salient object detection. We further apply the proposed method to selective image rendering and achieve better performance over the relevant existing algorithm.
    Journal of Visual Communication and Image Representation 01/2013; 25(3). DOI:10.1016/j.jvcir.2013.11.009 · 1.36 Impact Factor
  • Lu Dong, Weisi Lin, Chenwei Deng, Ce Zhu, Hock Soon Seah
    [Show abstract] [Hide abstract]
    ABSTRACT: For high-quality image rendering using Monte Carlo methods, a large number of samples are required to be computed for each pixel. Adaptive sampling aims to decrease the total number of samples by concentrating samples on difficult regions. However, existing adaptive sampling schemes haven't fully exploited the potential of image regions with complex structures to the reduction of sample numbers. To solve this problem, we propose to exploit uncertainty masking in adaptive sampling. Experimental results show that incorporation of uncertainty information leads to significant sample reduction and therefore time-savings.
    Circuits and Systems (ISCAS), 2013 IEEE International Symposium on; 01/2013
  • Wei Ming Chiew, Feng Lin, Kemao Qian, Hock Soon Seah
    [Show abstract] [Hide abstract]
    ABSTRACT: Modern microscopic volumetric imaging processes lack capturing flexibility and are inconvenient to operate. Additionally, the quality of acquired data could not be assessed immediately during imaging due to the lack of a coherent real-time visualization system. Thus, to eliminate the requisition of close user supervision while providing real-time 3D visualization alongside imaging, we propose and describe an innovative approach to integrate imaging and visualization into a single pipeline called an online incrementally accumulated rendering system. This system is composed of an electronic controller for progressive acquisition, a memory allocator for memory isolation, an efficient memory organization scheme, a compositing scheme to render accumulated datasets, and accumulative frame buffers for displaying non-conflicting outputs. We implement this design using a laser scanning confocal endomicroscope, interfaced with an FPGA prototyping board through a custom hardware circuit. Empirical results from practical implementations deployed in a cancer research center are presented in this paper.
    Computers in Industry 01/2013; DOI:10.1016/j.compind.2013.10.002 · 1.46 Impact Factor
  • [Show abstract] [Hide abstract]
    ABSTRACT: In this work, we propose a method to detect visually salient objects in computer synthesized images from 3D meshes. Different from existing detection methods on graphic saliency which compute saliency based on pixel-level contrast, the proposed method computes saliency by measuring object-level contrast of each object to the other objects in a rendered image. Given a synthesized image, the proposed method first extracts dominant colors from each object, and represents each object with the dominant color descriptor (DCD). Saliency is measured as the contrast between the DCD of the object and the DCDs of its surrounding objects. We evaluate the proposed method on a data set of computer rendered images, and the results show that the proposed method obtains much better performance compared with existing related methods.
    Visual Communications and Image Processing (VCIP), 2013; 01/2013
  • Zheng Zhang, Hock Soon Seah
    [Show abstract] [Hide abstract]
    ABSTRACT: Tracking of 3D human body movement from multiple camera video streams is an important problem in the domain of computer vision. In this paper we perform body pose tracking in 3D space using 3D data reconstructed at every frame. We present an efficient GPU-based method for 3D reconstruction of the real world dynamic scenes. Besides volumetric reconstruction, we propose to compute view-independent 3D optical flow (i.e., scene flow) in combination with volumetric reconstruction, and have attained efficient scene flow estimation using GPU acceleration. Body pose estimation starts from a deterministic prediction based on scene flow, and then uses a multi-layer search algorithm involving stochastic search and local optimization. We design and parallelize the PSO-based (particle swarm optimization) stochastic search algorithm and 3D DT (distance transform) computation of the pose estimation method on GPU. To the end, our system can reach efficient and robust body pose tracking.
    Proceedings of the 2012 IEEE 18th International Conference on Parallel and Distributed Systems; 12/2012
  • Wei Ming Chiew, Feng Lin, Kemao Qian, Hock Soon Seah
    [Show abstract] [Hide abstract]
    ABSTRACT: Non-rigid registration is crucial in imaging, in particular, to adjust deformities produced during image acquisition and improve the accuracy of datasets. However, conventional imaging systems lack the desired speed and computational bandwidth for additional non-rigid registration of the deformed images. Therefore, such functionality is usually unavailable in time-critical settings. Expensive computations and memory intensive characteristics of non-rigid image registration algorithms such as the Demons algorithm further limits the realization of such systems. In response, we propose an alternative and efficient custom hardware-based Demons registration algorithm which utilizes pipelined streaming models to minimize memory fetches for computation. Designed for highly customizable hardware, our design only requires single-pass of images to compute the Demons kernel. Implementation results on the Xilinx ML605 FPGA system is presented and quantitatively evaluated in clock cycle counts in contrast with a software-based implementation.
    Proceedings of the 2012 IEEE 14th International Conference on High Performance Computing and Communication & 2012 IEEE 9th International Conference on Embedded Software and Systems; 06/2012
  • Jun Yu, Dongquan Liu, Dacheng Tao, Hock Soon Seah
    [Show abstract] [Hide abstract]
    ABSTRACT: How do we retrieve cartoon characters accurately? Or how to synthesize new cartoon clips smoothly and efficiently from the cartoon library? Both questions are important for animators and cartoon enthusiasts to design and create new cartoons by utilizing existing cartoon materials. The first key issue to answer those questions is to find a proper representation that describes the cartoon character effectively. In this paper, we consider multiple features from different views, i.e., color histogram, Hausdorff edge feature, and skeleton feature, to represent cartoon characters with different colors, shapes, and gestures. Each visual feature reflects a unique characteristic of a cartoon character, and they are complementary to each other for retrieval and synthesis. However, how to combine the three visual features is the second key issue of our application. By simply concatenating them into a long vector, it will end up with the so-called "curse of dimensionality," let alone their heterogeneity embedded in different visual feature spaces. Here, we introduce a semisupervised multiview subspace learning (semi-MSL) algorithm, to encode different features in a unified space. Specifically, under the patch alignment framework, semi-MSL uses the discriminative information from labeled cartoon characters in the construction of local patches where the manifold structure revealed by unlabeled cartoon characters is utilized to capture the geometric distribution. The experimental evaluations based on both cartoon character retrieval and clip synthesis demonstrate the effectiveness of the proposed method for cartoon application. Moreover, additional results of content-based image retrieval on benchmark data suggest the generality of semi-MSL for other applications.
    IEEE transactions on systems, man, and cybernetics. Part B, Cybernetics: a publication of the IEEE Systems, Man, and Cybernetics Society 04/2012; 42(5):1413-27. DOI:10.1109/TSMCB.2012.2192108 · 3.78 Impact Factor
  • Minglei Liu, Hock Soon Seah, Ce Zhu, Weisi Lin, Feng Tian
    [Show abstract] [Hide abstract]
    ABSTRACT: In this paper, we present a reversible data embedding scheme based on an adaptive edge-directed prediction for images. It is known that the difference expansion is an efficient data embedding method. Since the expansion on a large difference will cause a significant embedding distortion, a location map is usually employed to select small differences for expansion and to avoid overflow/underflow problems caused by expansion. However, location map bits lower payload capacity for data embedding. To reduce the location map, our proposed scheme aims to predict small prediction errors for expansion by using an edge detector. Moreover, to generate a small prediction error for each pixel, an adaptive edge-directed prediction is employed which adapts reasonably well between smooth regions and edge areas. Experimental results show that our proposed data embedding scheme for natural images can achieve a high embedding capacity while keeping the embedding distortion low.
    Signal Processing 03/2012; 92(3):819-828. DOI:10.1016/j.sigpro.2011.09.028 · 2.24 Impact Factor