Ya-Ping Wong

Improved Canny Edges Using Ant Colony Optimization

Ant colony optimization (ACO) is a metaheuristic approach for solving hard optimization problem. It has been applied to solve various image processing problems such as image segmentation, classification, image analysis and edge detection. In this paper, we present an Improved Canny edges (ICE-ACO) algorithm which uses ACO to solve the problem of linking disjointed edges produced by Canny edge detector.

Facial Expression Recognition for E-learning Systems using Gabor Wavelet a Neural Network

In this paper, we present our initial studies and results obtained on e-learning facial expression recognition using Gabor Wavelet for facial feature extraction, and Back-propagation Neural Network for expression classification. An eFEC database that consists of 600 facial expression images is built for our research. We also provide reasons on why we need to build our own database instead of customizing and use current available expressions database, and what are the differences between eFEC databases compared with others. You may find comparisons in various experiment results presented. We believe the information would be very useful as to provide guideline and direction on how to improve the system performance and make it applicable in real-life elearning environments. (Abbreviation: eFEC --learning Facial Expression Classification; cca - correct classification in average)

Image Texture Classification Using Combined Grey Level Co-Occurrence Probabilities and Support Vector Machines

Texture refers to properties that represent the surface or structure of an object and is defined as something consisting of mutually related elements. The main focus in this study is to do texture segmentation and classification for texture digital images. Grey level co-occurrence probabilities (GLCP) method is being used to extract features from texture image. Gaussian support vector machines (GSVM) have been proposed to do classification on the extracted features. A popular Brodatz texture album had been chosen to test out the result. In this study, a combined GLCP-GSVM shows an improvement over GLCP in terms of classification accuracy.

A choice of weights for convex combination methods in estimating partial derivatives

Weights in convex combination methods are important parameters in the estimation of partial derivative values in three dimensional scattered data points. We introduce the three angles of the base triangle as weights in order to have more options for weight variety.

Vision-Based Hand Detection for Registration of Virtual Objects in Augmented Reality

This paper presents a markerless Augmented Reality (AR) framework which utilizes outstretched hand for registration of virtual objects in the real environment. Hand detection methods used in this work are purely based on computer vision algorithms to detect bare hand without assistance from markers or any other devices such as mechanical devices and magnetic devices. We use a stereo camera to capture video images, so that the depth information of the hand can be constructed. Skin color segmentation is employed to segment the hand region from the images. Instead of fiducial markers, the center of the palm and fingertips are tracked in real time. Incorporating the depth information computed, the 3D positions of these hand features are then used to estimate the 6DOF camera pose with respect to the hand, which in turn allows the virtual objects to be augmented onto the palm accurately. This method increases the ease of manipulation of AR objects. Users can inspect and manipulate the AR objects in an intuitive way by using their bare hands.

Facial expression analysis in e-learning systems - the problems and feasibility

Despite increasing so called robust algorithms have been proposed, classifying facial expressions in real-time systems, especially in e-learning, is a different story and not an easy job. Some influential aspects such as machine and hardware limitation, environmental issues, and even suitable expressions (from students) categories for e-learning systems are often neglected. In this paper, we discuss extensively on these problematic yet neglected issues and study for the possible, existing methods to handle them for an e-learning system. For some special cases, we suggest for alternatives other than facial expressions analysis, which may help the system to perform better in terms of user modeling. The paper is concluded with suggestion for more advanced and feasible analysis of students other behaviors (despite facial expressions) with the idea that no additional input device should be put on students body, which can cause uneasy feel on them.

Adaptive Model Simplification in Real-Time Rendering for Visualization

In this paper, we propose four different geometric measures to identify appropriate triangles to be simplified in 3D complex model. Each measure yields different weight on the same surface and produces a unique simplified model that worth to be analyzed. The proposed measures involve consideration on the resulting of the surfaces collapse, the high peak and low peak of the triangles mesh, the irregular triangle shape, the capacity and boundary view on the triangles mesh. The chosen triangle is to be collapsed based criterion on Half-edge Collapse Transformation method. From the empirical results, one of the proposed measures presents almost excellence in all the criteria mentioned above. The empirical results include the quality of the surface models (visualization purpose), the efficiency of the measures and the overall appearance preservation of the simplified models. The proposed measures are then to be compared to three existing measures. From the analyzed results, we combine the measures to adapt to the user’s response for generating the user-desired simplified models.

Improvement in decimation of triangle meshes for level of detail

Simplifying polygonal models to achieve a constant frame rate or to generate an ideal size of an object proportional to its viewing distance is one of the many techniques used in 3D visualizations these days. In this paper, we have selected a decimation algorithm to be further enhanced by introducing other techniques to obtain a better output. In our technique, parts of the characterization vertices are identified and further analyzed. In the boundary convex group, categorization is made whether the vertex is to be deleted or preserved. In the evaluation stage of the decimation algorithm, we use the singular value decomposition algorithm to compute the smallest eigenvector from a matrix formed from the surrounding neighboring vertices of the simple candidate vertices. Finally, in the triangulation stage, a careful and simple patching step is applied to the resulting holes so that the output would be balance in sizes. A balanced size refers to re-generating triangle-strips of similar size of the edges for a smoother model viewing.

Density-Based Denoising of Point Cloud

Point cloud source data for surface reconstruction is usually contaminated with noise and outliers. To overcome this deficiency, a density-based point cloud denoising method is presented to remove outliers and noisy points. First, particle-swam optimization technique is employed for automatically approximating optimal bandwidth of multivariate kernel density estimation to ensure the robust performance of density estimation. Then, mean-shift based clustering technique is used to remove outliers through a thresholding scheme. After removing outliers from the point cloud, bilateral mesh filtering is applied to smooth the remaining points. The experimental results show that this approach, comparably, is robust and efficient.

Cluster Coding Algorithm for Stochastic Textures Identification

A stochastic texture is a texture whereby the arrangement of the pattern is random in nature. The identification for each of these textures is uncertain and usually involves complex methods to search for the locations of the correlation between these textures. We propose the cluster coding algorithm which could easily classify the stochastic textures in a semantic way based on statistical features. This algorithm is successfully applied in both synthetic and real-life textures for segmentation. In this study, the cluster coding showed a significant improvement over other techniques in terms of classification accuracy and computation time.