Paul Kwok

Sequential approach to blind source separation using second order statistics

A general result on identifiability for the blind source separation problem, based on second order statistics only, is presented. The separation principle using second order statistics is first proposed. This is followed by a discussion on a number of algorithms to separate the sources one by one.

Modeling the optical constants of AlxGa1−xAs alloys

The extension of Adachis model with a Gaussian-like broadening function, in place of Lorentzian, is used to model the optical dielectric function of the alloy AlxGa1-xAs. Gaussian-like broadening is accomplished by replacing the damping constant in the Lorentzian line shape with a frequency dependent expression. In this way, the comparative simplicity of the analytic formulas of the model is preserved, while the accuracy becomes comparable to that of more intricate models, and/or models with significantly more parameters. The employed model accurately describes the optical dielectric function in the spectral range from 1.5 to 6.0 eV within the entire alloy composition range. The relative rms error obtained for the refractive index is below 2.2% for all compositions

Thyroid cancer cells boundary location by a fuzzy edge detection method

Morphometric assessment of tumor cells is important in the prediction of biological behavior of thyroid cancer. In order to automate the process, the computer-based system has to recognize the boundary of the cells. Many methods for the boundary detection have appeared in the literature and some of them applied to microscopic slice analysis. However, there is no reliable method since the gray-levels in the nuclei are uneven and are similar to the background. In the paper, a fuzzy edge detection method is used and is based on an improved generalized fuzzy operator. The method enhances the nuclei and effectively separates the cells from the background.

Uncorrelated component analysis for blind source separation

The uncorrelated component analysis (UCA) of a stationary random vector process consists of searching for a linear transformation that minimizes the temporal correlation between its components. Through a general analysis we show that under practically reasonable and mild conditions UCA is a solution for blind source separation. The theorems proposed in this paper for UCA provide useful insights for developing practical algorithms. UCA explores the temporal information of the signals, whereas independent component analysis (ICA) explores the spatial information; thus UCA can be applied for source separation in some cases where ICA cannot. For blind source separation, combining ICA and UCA may give improved performance because more information can be utilized. The concept of single UCA (SUCA) is also proposed, which leads to sequential source separation.

Brain tumor boundary detection in MR image with generalized fuzzy operator

Boundary detection in MR image with brain tumor is an important image processing technique applied in radiology for 3D reconstruction. The nonhomogeneities density tissue of the brain with tumor can result in achieving the inaccurate location in any boundary detection algorithms. Recently, some studies using the contour deformable model with regional base technique, the performance is insufficient to obtain the fine edge in the tumor, and the considerable error in accuracy is existed. Moreover, even in some of the normal tissue region, edge created by this method has also been encompassed. In this paper, we propose a new approach to detect the boundary of brain tumor based on the generalized fuzzy operator (GFO). One typical example is used for evaluating this method with the contour deformable model.

Modeling the optical constants of GaP, InP, and InAs

An extension of the Adachi model with the adjustable broadening function, instead of the Lorentzian one, is employed to model the optical constants of GaP, InP, and InAs. Adjustable broadening is modeled by replacing the damping constant with the frequency-dependent expression. The improved flexibility of the model enables achieving an excellent agreement with the experimental data. The relative rms errors obtained for the refractive index equal 1.2% for GaP, 1.0% for InP, and 1.6% for InAs.

Normalization of contrast in document images using generalized fuzzy operator with least square method

The visual effect of non-uniform contrast and brightness surrounds in the image is a very common problem in the applications of photocopying, IC manufacture and medicine. In using the digital/CCD camera to capture documents and photos based on non-uniform illumination condition, the poor image will be seen. The poor image can result in achieving the inaccurate reading from the optical character recognition (OCR) system. This paper presents a new approach to normalize the local contrast in documentation based on the least squares method and also enhance the object of interest using the generalized fuzzy operator (GFO). Two typical examples are used for evaluating the method.

Estimation of the gray level variations in soft and hard peri-implant tissue from X-ray images

In the treatment and assessment of periodontal disease periapical films are taken at regular intervals. The films are compared around the region of the implant to monitor the changes that have taken place. A computer-assisted method to automate this process is presented here. To compare two images, a pair of X-ray images are superimposed on top of each other. The images are normalized and transformed to obtain the best fit. These images are subtracted and the difference in pixel values are used as a basis for analysis. A method for improving the alignment accuracy and contrast compensation is highlighted.

Blind separation and localization of dipole sources of MEG

We present a new approach to MEG inverse problem by modeling it into a standard blind source separation problem. In our approach, dipole sources and gain matrix are estimated without any knowledge about the head geometry and conductivity. Given the head model, we can compute dipole locations further. Our matrix pencil method developed before is suitable for this task and is applied in the simulation. Simulation results are presented.

Multi-channel Blind Blur Identification and Image Restoration

This paper considers the problem of multi-channel blind image restoration and blur identification. By constructing the blind identification problem into an optimization problem, we propose a subspace decomposition based algorithm to blindly identify the blur functions. The proposed algorithm is inherently the same as many of the others in the literature, but at significantly reduced computation complexity. Let M be the number of blurred images available, N1 X N2 be the size of the images and L1 X L2 be the size of blur functions, our algorithm has a computational complexity of O(M2L21L22N1N2), as compared to O(M4L21L22N1N2) for previous works. The proposed algorithm is therefore more suitable for practical applications.