Research

Research Interests 

•  Computer Vision
• Object Recognition
• Motion Estimation
• Image Inpainting
• Tracking
• Computer Graphics
  
• Image Processing
• Machine Learning

Description: In this work, we propose a method for extracting salient local features from 3D models using shape variation which has application to 3D surface registration. In the proposed technique, the surface shape at a point is specified by a quantitative measure known as the shape index. It is invariant to rigid transformations such as translation and rotation. The shape index at a point is calculated at multiple scales by fitting a surface to the local neighbourhoods of different sizes. The local surface variation is then measured by calculating the variation of the shape index of every point in the neighbourhood. Points corresponding to local maxima of surface variation are selected as suitable features. Registration results using our feature extraction framework and spin images as local descriptors have demonstrated the effectiveness of the approach.

Results:
                   

Features extracted from the surfaces and registration results using spin images. Positions marked in red are the scene scan aligned w.r.t the model scan.

References:

H.T. Ho and D. Gibbins, "Multi-scale Feature Extraction for 3D Surface Registration using Local Shape Variation", IVCNZ 2008.

Description: In this work, we present a method for extracting salient local features from 3D models using surface curvature which has application  to 3D object recognition. In the developed technique, the amount of curvature at a point is specified by a positive number known as the curvedness. This value is invariant to rotation as well as translation. A local description of the surface is generated by fitting a surface to the neighbourhood of a keypoint and estimating its curvedness at multiple scales. From this surface, points corresponding  to local maxima and minima of curvedness are selected as suitable features and a confidence measure of each keypoint is also calculated based on the deviation of its curvedness from the neighbouring values.

Results:

                         

Features extracted from different models

References:

H.T. Ho and D. Gibbins, "Multi-scale Feature Extraction from 3D Models using Local Surface Curvature", DICTA 2008.

Description: In this work, we propose a fully automatic framework for efficient image completion with quaternion potential functions. Our framework extends a new exemplar-based algorithm  which uses a discrete global optimization strategy based on Markov random fields (MRFs) and belief propagation (BP). To efficiently optimize the MRF, this method combines the priority belief propagation and dynamic label pruning to reduce the computational cost while still produce high quality results. However, one of the drawbacks of the technique is its use of a heuristically chosen parameter set. A method for automatically determining the parameters for the belief propagation and dynamic label pruning steps is presented. In order to estimate the optimal parameters of the algorithm, we employ an information theoretic approach making use of the entropy of the image patches and the distribution of pairwise node potentials. Furthermore, we also introduce the use of quaternions or hypercomplex numbers in estimating the potential functions to better capture the correlations across the color layers of the input image. By correlating quaternion image patches based on the recently developed concepts of quaternion Fourier transforms and quaternion correlation for color images, the potential functions can be calculated efficiently.
 
Results:

Completion results for different images

References:

H.T. Ho and R. Goecke, "An Automatic Framework for Efficient Image Completion using Markov Random Fields with Quaternion Potential Functions", submitted to the IEEE Transactions on Image Processing.
H.T. Ho and R. Goecke, “Quaternion Potential Functions for a Color Image Completion Method using Markov Random Fields”,  DICTA 2007.
H.T. Ho and R. Goecke, “Automatic Parametrisation for an  Image Completion Method based on Markov Random Fields”, ICIP 2007.

Optical Flow Estimation using Fourier Mellin Transforms (FMT)

Description: In this work, we propose a novel method of computing the optical flow using the Fourier Mellin Transform (FMT). Each image in a sequence is divided into a regular grid of patches and the optical flow is estimated by calculating the phase correlation of each pair of co-sited patches using the FMT. By applying the FMT in calculating the phase correlation, we are able to estimate not only the pure translation, as limited in the case of the basic phase correlation techniques, but also the scale and rotation motion of image patches, i.e. full similarity transforms. Moreover, the motion parameters of each patch can be estimated to sub-pixel accuracy based on a recently proposed algorithm that uses a 2D esinc function in fitting the data from the phase correlation output. We also improve the estimation of the optical flow by presenting a method of smoothing the field by using a vector weighted average filter.

Results:

A frame of the Venus sequence and its log polar transfrom

Comparison of different approaches