Kwok-Leung Chan

A probabilistic relaxation labeling framework for reducing the noise effect in geometric biclustering of gene expression data

Biclustering is an important method in DNA microarray analysis which can be applied when only a subset of genes is co-expressed in a subset of conditions. Unlike standard clustering analyses, biclustering methodology can perform simultaneous classification on two dimensions of genes and conditions in a microarray data matrix. However, the performance of biclustering algorithms is affected by the inherent noise in data, types of biclusters and computational complexity. In this paper, we present a geometric biclustering method based on the Hough transform and the relaxation labeling technique. Unlike many existing biclustering algorithms, we first consider the biclustering patterns through geometric interpretation. Such a perspective makes it possible to unify the formulation of different types of biclusters as hyperplanes in spatial space and facilitates the use of a generic plane finding algorithm for bicluster detection. In our algorithm, the Hough transform is employed for hyperplane detection in sub-spaces to reduce the computational complexity. Then sub-biclusters are combined into larger ones under the probabilistic relaxation labeling framework. Our simulation studies demonstrate the robustness of the algorithm against noise and outliers. In addition, our method is able to extract biologically meaningful biclusters from real microarray gene expression data.

Fractal dimension estimation and noise filtering using Hough transform

This paper considers the scenario of a periodic signal consisting of a sum of sinusoids with different unknown frequencies and phases superimposed on fractal noise of an unknown fractal dimension. A novel method is reported to estimate the fractal dimension of the noise. The problem is first converted to a line detection problem. Then the Hough transform is applied. A comparison with the least-squares estimation method is reported. It is shown that the method detects the fractal dimension more accurately and robustly. Using the estimated fractal dimension, the Wiener filter is used to remove the fractal noise. Its performance is also compared with some other common filters. Finally, an indicator is suggested to estimate the likelihood that the input signal belongs to the desired signal model, i.e. a periodic signal with superimposed fractal noise.

Multivariate hierarchical Bayesian model for differential gene expression analysis in microarray experiments

BackgroundIdentification of differentially expressed genes is a typical objective when analyzing gene expression data. Recently, Bayesian hierarchical models have become increasingly popular to solve this type of problems. These models show good performance in accommodating noise, variability and low replication of microarray data. However, the correlation between different fluorescent signals measured from a gene spot is ignored, which can diversely affect the data analysis step. In fact, the intensities of the two signals are significantly correlated across samples. The larger the log-transformed intensities are, the smaller the correlation is.ResultsMotivated by the complicated error relations in microarray data, we propose a multivariate hierarchical Bayesian framework for data analysis in the replicated microarray experiments. Gene expression data are modelled by a multivariate normal distribution, parameterized by the corresponding mean vectors and covariance matrixes with a conjugate prior distribution. Within the Bayesian framework, a generalized likelihood ratio test (GLRT) is also developed to infer the gene expression patterns. Simulation studies show that the proposed approach presents better operating characteristics and lower false discovery rate (FDR) than existing methods, especially when the correlation coefficient is large. The approach is illustrated with two examples of microarray analysis. The proposed method successfully detects significant genes closely related to the experimental states, which are verified by the biological information.ConclusionsThe multivariate Bayesian model, compatible with the dependence between mean and variance in the univariate Bayesian model, relaxes the constant coefficient of variation assumption between measurements by adding a covariance structure. This model improves the identification of differentially expressed genes significantly since the Bayesian model fit well with the microarray data.

Volumetric Model Reconstruction from Unrestricted Camera Views Based on the Photo-consistency of 3D Voxel Mask

This project aims to develop a three-dimensional (3D) model reconstruction system using images acquired from a mobile camera. It consists of four major steps: camera calibration, volumetric model reconstruction, surface modeling and texture mapping. A novel online scale factor estimation is developed to enhance the accuracy of the coplanar camera calibration. For the volumetric modeling, the voting-based shape-from-silhouette first generates a coarse model, which is then refined by the photo-consistency check using the novel 3D voxel mask. Our scheme can handle concave surface in a sophisticated way. Finally, the surface model is formed with the original images mapped. 3D models of some test objects are presented.

Detection of swimmer using dense optical flow motion map and intensity information

A vision-based system that can locate individual swimmers and recognize the activities is applicable for swimming gait analysis, drowning event detection, etc. The system relies on accurate detection of swimmer’s body parts such as head and upper limbs. The swimmer detection problem can be regarded as background subtraction. Swimmer detection in the aquatic environment is very difficult due to a dynamic background with water ripples, splashes, specular reflections, etc. This paper presents a swimmer detection method which utilizes both local motion and intensity information estimated from the image sequence. Local motion information is obtained by computing dense optical flow and periodogram. We adopt a heuristic approach to generate a motion map characterizing the local motions (random/stationary, ripple or swimming) of image pixels over a short duration. Intensity information is modeled as a mixture of Gaussians. Finally, using the motion map and the Gaussian models, swimmers are detected in each video frame. We test the method on video sequences captured at daytime, and nighttime, and of different swimming styles (breaststroke, freestyle, backstroke). Our method can detect swimmers much better than that using intensity information alone. In addition, we compare our method with existing algorithms—codebook model and self-organizing artificial neural networks. The methods are tested on publicly available video sequence and our swimming video sequence. We show through the quantitative measures the superiority of our method.

Object-Based Image Retrieval Using Hierarchical Shape Descriptor

Shape is the most basic and convenient feature to describe objects. Retrieval by shape similarity is implemented in this project. Object shapes are segmented into tokens according to their local feature of minimum turn angle. User sketch is the query input and the retrieval algorithm matches the sketch with the nearest object in the database by using features distance. Scaling, rotation and missing sketch of objects are also considered in this paper. Together with the M-tree indexing, the system performance can be strengthened. However, many objects have similar outer shape boundary but different inner shapes. The retrieval accuracy will be affected by this situation. Hierarchical Shape Descriptor is proposed to solve the problem. It can distinguish similar outer boundaries but with different inner shapes objects. A completely new image retrieval system is implemented in order to accommodate the new image content descriptor. Our results show that the proposed system is fairly accurate and the Hierarchical Shape Descriptor is a better image content descriptor than the existing method using only the outer boundary.

3D object model reconstruction from image sequence based on photometric consistency in volume space

This paper presents a system that can reconstruct a photorealistic 3D object model from an image sequence captured at arbitrary viewpoints. The whole system consists of four steps: camera calibration, volumetric modeling, polygonal model formation and texture mapping. We adopt the shape-from-silhouette approach for volumetric modeling. There are two common types of object surface that are difficult to reconstruct—textureless surface and concave surface. To tackle the problems, we propose to perform the volumetric modeling based on the constraints of viewpoint proximity and photometric consistency in the volume space. The volumetric model is converted to the mesh model for efficient manipulation. Finally, the texture map is generated from the image sequence to give the 3D model a photorealistic appearance. Some reconstructed object models are presented to demonstrate the superior performance of our system as compared with the conventional modeling technique based on the photo-consistency in the image space.

Model-based human motion analysis in monocular video

Tracking human motion in monocular video is a challenging problem in computer vision. It has found a wide range of applications, such as visual surveillance, virtual reality, sports science, etc. This project aims to develop a model-based human motion analysis system that can track human movement in a monocular image sequence with minimum constraint. No markers or sensors are attached to the subject. Given a video clip, the first step is to fit the 3D human model manually to the subject in the first frame of the video. Then background subtraction is used to extract the human silhouette. We propose the silhouette chamfer as the main matching feature. A chamfer distance measure is carried out on the extracted subject silhouette. The silhouette chamfer contains both the chamfer distance and region information. Finally, we use a discrete Kalman filter to predict the pose of the subject in each image frame. The updating step uses Broydents method to optimize the predicted pose to fit the persons silhouette by using the cost function. We use the gait database SOTON to test our system. The image sequences contain human walking in both indoor and outdoor environments. The motion tracking results demonstrate that our system has an encouraging performance.

Detection of foreground in dynamic scene via two-step background subtraction

Various computer vision applications such as video surveillance and gait analysis have to perform human detection. This is usually done via background modeling and subtraction. It is a challenging problem when the image sequence captures the human activities in a dynamic scene. This paper presents a method for foreground detection via a two-step background subtraction. Background frame is first generated from the initial image frames of the image sequence and continuously updated based on the background subtraction results. The background is modeled as non-overlapping blocks of background frame pixel colors. In the first step of background subtraction, the current image frame is compared with the background model via a similarity measure. The potential foregrounds are separated from the static background and most of the dynamic background pixels. In the second step, if a potential foreground is sufficiently large, the enclosing region is compared with the background model again to obtain a refined shape of the foreground. We compare our method with various existing background subtraction methods using image sequences containing dynamic background elements such as trees and water. We show through the quantitative measures the superiority of our method.

Removal of Specular Reflection Component Using Multi-view Images and 3D Object Model

Image-based 3D model reconstruction method can use the same multi-view image sequence of the object for the generation of both the geometry model and texture map. Texture is very critical for virtual exhibition of 3D model and should be of high quality comparable to the geometry data. One problem is that the object surface may exhibit specular reflection of illuminated light. The texture extracted directly from the images can be unnatural. We propose a method for the removal of specular reflection component in each image. Each camera view is calibrated and a 3D mesh model of the object is generated. For each triangle patch, the projected colors on all visible views are found. The specular chromaticity is replaced by the corresponding diffuse chromaticity. We test the method on image sequences of synthetic and real objects. The diffuse image sequence can be used to generate the texture map.