Aftab Khan

Efficient blind image deconvolution using spectral non-Gaussianity

The principle of Independent Component Analysis ICA has been used in blind signal separation and deconvolution problems. In image restoration, such methods are often computationally intensive and ringing and noise amplification artifacts from the deblurring process greatly affect the image statistics and vary the calculated non-Gaussianity measures. To overcome the problems, we propose an enhanced scheme that employs the non-Gaussianity principle of ICA on the spectrum rather than the image data itself. That is, the spectral kurtosis is used as a measure of non-Gaussianity during the deblurring process. The deblurring process measures the non-Gaussianity of the image spectrum of the estimated images and the value maximizes at the true blurring kernel. The optimal solution is sought through a Genetic Algorithm. The scheme is simple and efficient and does not require any prior knowledge about the image or the blurring process. Validations have been carried out on various examples and they show that spectral non-Gaussianity optimizes on the parameters in a close vicinity of the original blurring functions. Results are presented for both benchmark and real images. The proposed method achieves marked improved results over the existing methods.

The characterisation and modelling of manufacturing porosity of a 2-D carbon/carbon composite

The thermal properties of carbon/carbon composite are strongly affected by manufacturing porosity. This paper focuses on the characterisation of the manufacturing macro-porosity of a 2D carbon/carbon composite using X-ray computed tomography. The different types of manufacturing porosity were classified and quantified according to their size and location. Three types of macro-porosity were identified using computed tomography, namely trans-tow cracks, interfacial cracks and dry zones. A composite unit cell representing the three types of porosity was developed to model and investigate the effective transverse thermal transport properties (thermal conductivity and diffusivity) of the carbon/carbon composite. Finite element simulations and theoretical calculations were performed and compared with laser flash tests for validation. The influence of the stacking sequence of the laminates on porosity distribution and transverse thermal conductivity of the carbon/carbon composite was also investigated.

Spectral non-gaussianity for blind image deblurring

A blind image deblurring method based on a new nongaussianity measure and independent component analysis is presented. The scheme assumes independency among source signals (image and filter function) in the frequency domain. According to the Central Limit Theorem the blurred image becomes more Gaussian. The original image is assumed to be non-gaussian and using a spectral non-gaussianity measure (kurtosis or negentropy) one can estimate an inverse filter function that maximizes the non-gaussianity of the deblurred image. A genetic algorithm (GA) optimizing the kurtosis in the frequency domain is used for the deblurring process. Experimental results are presented and compared with some existing methods. The results show that the deblurring from the spectral domain offers several advantages over that from the spatial domain.

Quality measures for blind image deblurring

Blind image deblurring is limited by the unavailability or in many cases little information about the PSF. If the PSF is estimated, then deblurring simplifies to just deconvolving the blurred image with the PSF using any conventional deblurring filter. We have recently proposed a blind deblurring scheme using kurtosis measures. The scheme is able to deblur degraded images of unknown types such as out-of-focus, motion or atmospheric turbulence. However, for blurred images whose original are unknown, it is impossible to measure the improvement, unlike in simulated blurring cases. In this paper, a way of measuring the quality improvement of the deblurring is suggested. The deblurred image is-reblurred by the estimated PSF and then the PSNR between the original blurred image and the re-blurred image is calculated as an indication of deblurring quality. Deblurring filters often produce noise and ringing artifacts in the deblurred image, which will be less severe when a candidate filter similar to the true PSF is used. This quality measures further enhance the blind deblurring scheme and has been tested on both synthetic and real blurred images.

A dialectical analysis of non-reference image quality measures (IQMs) and restoration filters for single image blind deblurring

Blind image deblurring relies on a good estimation of Point Spread Function (PSF) and the utilization of an effective restoration filter. Even if the PSF is estimated well, the deblurring result depends heavily on the abilities of the restoration filter to produce a good approximation of the pristine image. Blind Image Quality Measures (IQMs) that guide the deblurring algorithm are also dependent on the restored image data. This research work evaluated the performance of the restoration filters and the blind IQMs when the true PSF has been estimate dand presents the effectiveness of both for blind deblurring. Wiener, Richardson-Lucy and Total Variation deblurring filters and BRISQUE, NIQE, SSEQ, Curvelet QA (CQA) IQMs have been analysed. Simulations have been performed over a wide range of images various blurring types (Gaussian, out-of-focus and motion). The results show that TV deblurring filter in conjunction with CQA deliver a near estimate of the pristine image for the artifically blurred images. In the case of real blurred images, Wiener filter presents high quality deblurred images and both SSEQ and CQA depict high quality images.