Kok Cheong Wong

Ellipse detection based on symmetry

Abstract In this paper, a novel and generic approach is proposed for detecting ellipses from symmetric contours in a picture efficiently. The idea is based on the detection of the symmetric axes from contours in a Hough-based approach, so as to transform a high-dimensional problem into two two-dimensional ones. From the idea, the algorithms for detecting symmetric axes and ellipses are designed and tested. From the algorithm analyses, this method is practical and runs faster than the current-reported methods. By experiments, this method demonstrates highly effective and robust results.

Detection and localisation of reflectional and rotational symmetry under weak perspective projection

In this paper, a novel method is proposed for detecting and recovering the pose of a reflectional and rotational symmetry from a single weak perspective image. In our paradigm, an effective and novel Hough-based approach is proposed for extracting skew symmetry axes from edge contours. A refinement algorithm is developed for verifying and pruning away the implausible and spurious hypothesised symmetry axes. For each of the detected skew symmetry axes and its corresponding skew angle, a mathematical formulation is devised for evaluating the geometric relationships between an orthogonal junction formed by a symmetry axis and its perpendicular axis in an original plane and its corresponding skew angle in an image plane. The derived geometric formulation is integrated into an effective voting scheme for estimating the projective parameters of a reflectionally and rotationally symmetric planar shape using the information of the plausible skew symmetry axes. Experimental results are presented to verify the computational performance of the proposed paradigm. The practical significance and generality of the proposed method are manifold.

Generating Consistent Motion Transition via Decoupled Framespace Interpolation

The framespace interpolation algorithm abstracts motion sequences as 1D signals, and interpolates between them to create higher dimension signals, with weights drawn from a user specified curve in a bounded region. We reformulate the algorithm to achieve motion‐state based transition via dynamic warping of framespaces and automatic transition timing via framespace frequency interpolation. Basis motions displaying diverse coordination configurations between upper and lower body‐halves, cannot be consistently corresponded at a macro level. We address this problem here, through decoupled blending of these halves to achieve true consistency, and eliminate accumulated phase differences via cosine phase warp functions. This generalization enables interpolation of motions with diverse coordinations between the upper and lower bodies.

Skinning With Deformable Chunks

This paper presents a new technique to generate skin deformations for articulated virtual characters. Deformable chunks are utilized to represent the internal structures of a character, and they are automatically generated beneath the skin depending on user specified patches on the skin. Powered by finite element method, these chunks can deform in a controllable manner according to the skeleton posture. The deformations of the chunks lead to natural deformation of the skin of the character. Intuitive rigs are devised to facilitate the adjustments of deformation results. We have demonstrated that most of the visual characteristics of skin deformations generated by real muscle contractions can be accomplished by our method.

Constrained framespace interpolation

This paper aims to consistently blend different types of motions after establishing automatic correspondence between their salient features. Framespace interpolation is a consistent forward kinematics motion transition technique that uses weights from input spline curves to warp and blend motions. Its application has been limited to interactive interpolation of two or four cyclic motions. Though based on principles that minimize violations like sliding of supporting end-effectors, it does not guarantee slide-free motion. This paper extends the application of framespace interpolation to an unlimited chain of cyclic and acyclic motions, via an improved coordination warp and constraints on transition curves. Inverse kinematics has been seamlessly used to correct slide-violations on the fly. These extensions have opened up exciting possibilities in real-time cyclification, blending and concatenation of a wide variety of human motions.

Recognition of two dimensional objects based on a novel generalized Hough transform method

In this paper, we present a model-based recognition system for identifying and estimating the pose of two dimensional arbitrary shapes subject to Euclidean and similarity transformations. A novel and effective paradigm based on the modification of the generalized Hough transform (GHT) is presented. In contrast to the classical GHT and the existing Hough-based methods, the storage space and computational complexity of the proposed method are reduced significantly by using an efficient voting scheme in conjunction with invariant geometric features. In the framework of the proposed system, the classical four dimensional Hough space is casted to a two dimensional Hough space. For many existing Hough-based methods, a scaling factor bound must be pre-specified based on prior knowledge of the given scene. In contrast, no such scaling factor bound is required in our proposed paradigm. Extensive experimental results are presented to verify the performance merits of our recognition system.

Semantic representation and correspondence for state-based motion transition

Consistent transition algorithms preserve salient source motion features by establishing feature-based correspondence between motions and accordingly warping them before interpolation. These processes are commonly dubbed as preprocessing in motion transition literature. Current transition methods suffer from a lack of economical and generic preprocessing algorithms. Classical computer vision methods for human motion classification and correspondence are too computationally intensive for computer animation. The paper proposes an analytical framework that combines low-level kinematics analysis and high-level knowledge-based analysis to create states that provide coherent snapshots of body-parts active during the motion. These states are then corresponded via a globally optimal search tree algorithm. The framework proposed here is intuitive, controllable, and delivers results in near realtime. The validity and performance of the proposed system are tangibly proven with extensive experiments.

Recognition of polyhedral objects using triplets of projected spatial edges based on a single perspective image

Abstract A novel and intuitive mathematical formulation is proposed for the interpretation of the geometric relationships of a triple of spatial edges and their perspective projection forming image lines. No restriction is imposed on the configuration of the triple of spatial edges. An eighth-degree polynomial equation explicitly defined by the space angles between the corresponding three spatial edges in an object frame is derived. The crux of this representation is that the angular attributes of pairs of spatial edges are expressed in an object-independent coordinate system. An effective hypothesis generation scheme is proposed which can take advantage of the representation. In the proposed recognition framework, triples of connected spatial edges and trihedral vertices are employed as key features for model invocation, hypothesis generation and pose estimation. Extensive experiments are performed to verify the plausibility and robustness of the proposed recognition methods. In general, the success rate for identifying and localising polyhedral objects using the proposed recognition method is relatively high.

Dust and water splashing models for hopping figures

To enhance the realism of computer animation sequences, the interaction between autonomous creatures and their environment needs to be modelled. Two such commonly occurring phenomena, dust and water splashing on impact, are presented in this paper. These pseudodynamic models, based on particle system concepts, are designed for hoppers with approximately elliptical bases. The proposed models can be extended for use with multi-legged creatures. The dust model is parameterized by creature dimensions, soil particle dimensions and wind direction. The splash model is parameterized by creature dimensions and water film thickness. Explicit control of behavioural parameters of both models is also provided to allow application-specific usage. Copyright

NeuroEnveloping: a transferable character skin deformation technique

We proposed a character skin deformation method, namely neuroEnveloping. The main idea relies on employing artificial neural networks to learn how to synthesize deformable skin from example shapes paired with skeleton postures of a character. Furthermore, having encoded the deformation information in the neural networks, the skinning model can be applied for the deformation of other similar characters. The character skin is decomposed into patches according to a base control mesh which is introduced as a common platform in the proposed framework. The deformation of each patch of the character is controlled by a trained neural network. We devised an effective stitching operator to eliminate discontinuities between neighboring patches. The experimental results show that the proposed approach can generate aesthetically pleasing results at interactive speed.