Noiseless Image Research Articles

Data collection and processing

Data collection and processing

Bias error reduction of digital image correlation using Gaussian pre-filtering

In digital image correlation (DIC), the iterative spatial domain cross-correlation algorithm using high-order B-spline interpolation algorithms has been strongly recommended for accurate sub-pixel displacement measurement. However, the magnitude of the position-dependent bias error increases with the increase of noise level, which dramatically reduces the registration accuracy of DIC for real experimental images. In this paper, a simple method, based on pre-smoothing the speckle images with a 5×5pixels Gaussian low-pass filter prior to correlation analysis, is proposed for reducing the bias error in measured displacements. Both numerical simulations and real experiments reveal that the proposed technique is capable of reducing the bias error in measured displacement to a negligible degree for both noisy and noiseless images, even though a simple bicubic interpolation is used.

Guided image filtering using signal subspace projection

There are various image filtering approaches in computer vision and image processing that are effective for some types of noise, but they invariably make certain assumptions about the properties of the signal and/or noise which lack the generality for diverse image noise reduction. This study describes a novel generalised guided image filtering method with the reference image generated by signal subspace projection (SSP) technique. It adopts refined parallel analysis with Monte Carlo simulations to select the dimensionality of signal subspace in the patch-based noisy images. The noiseless image is reconstructed from the noisy image projected onto the significant eigenimages by component analysis. Training/test image are utilised to determine the relationship between the optimal parameter value and noise deviation that maximises the output peak signal-to-noise ratio (PSNR). The optimal parameters of the proposed algorithm can be automatically selected using noise deviation estimation based on the smallest singular value of the patch-based image by singular value decomposition (SVD). Finally, we present a quantitative and qualitative comparison of the proposed algorithm with the traditional guided filter and other state-of-the-art methods with respect to the choice of the image patch and neighbourhood window sizes.

Reliable Face Recognition Using Artificial Neural Network

Facial detection and recognition are among the most heavily researched fields of computer vision and image processing. Most of the current face recognition techniques suffer when the noises affect the global features or the local intensity pixels of the images under consideration. In the proposed Reliable Face Recognition System (RFRS) system, for the first time, a combination of Gabor Filter (GF), Principal component analysis (PCA) for efficient feature extraction, and ANN for classification is employed. This demonstrates how to detect faces in noisy images by training the network several times on various input; ideal and noisy images of faces. Applying GF before PCA reduces PCA sensitivity to noise, provides a greater level of invariance, and trains the ANN on different sets of noisy images. The output of the ANN is a vector whose length equal to the distinct subjects in Olivetti Research Laboratory (ORL). The ANN is trained to output a 1 in the correct position of the output vector and to fill the rest of the output vector with 0’s. Experimentation is carried out on RFRS by using ORL datasets. The experimental results show that training the network on noisy input images of face greatly reduce its errors when it had to classify or recognize noisy images. For noisy face images, the network did not make any errors for faces with noise of mean 0.00 or 0.05, while the average recognition rate varies from 96.8% to 98%. When noise of mean 0.10 is added to the images the network begins to make errors. For noiseless face images, the proposed system achieves correct classification. Performance comparison between RFRS and other face recognition techniques shows that for most of the cases, RFRS performs better than other conventional techniques under different types of noises and it shows the high robustness of the proposed algorithm.

EXCALIBUR: a small-pixel photon counting area detector for coherent X-ray diffraction - Front-end design, fabrication and characterisation

Coherent X-ray diffraction experiments on synchrotron X-ray beamlines require detectors with high spatial resolution and large detection area. The read-out chip developed by the MEDIPIX3 collaboration offers a small pixel size of 55 microns resulting in a very high spatial resolution when coupled to a direct X-ray conversion segmented silicon sensor. MEDIPIX3 assemblies present also the advantages of hybrid pixel detectors working in single photon counting mode: noiseless imaging, large dynamic range, extremely high frame rate. The EXCALIBUR detector is under development for the X-ray Coherence and Imaging Beamline I13 of the Diamond Light Source. This new detector consists of three modules, each with 16 MEDIPIX3 chips which can be read-out at 100 frames per second in continuous mode or 1000 frames per second in burst mode. In each module, the sensor is a large single silicon die covering 2 rows of 8 individual MEDIPIX3 read-out chips and provides a continuous active detection region within a module. Each module includes 1 million solder bumps connecting the 55 microns pixels of the silicon sensor to the 55 microns pixels of the 16 MEDIPIX3 read-out chips. The detection area of the 3-module EXCALIBUR detector is 115 mm × 100 mm with a small 6.8 mm wide inactive region between modules. Each detector module is connected to 2 FPGA read-out boards via a flexi-rigid circuit to allow a fully parallel read-out of the 16 MEDIPIX3 chips. The 6 FPGA read-out boards used in the EXCALIBUR detector are interfaced to 6 computing nodes via 10Gbit/s fibre-optic links to maintain the very high frame-rate capability. The standard suite of EPICS control software is used to operate the detector and to integrate it with the Diamond Light Source beamline software environment. This article describes the design, fabrication and characterisation of the MEDIPIX3-based modules composing the EXCALIBUR detector.

Strategies for the deconvolution of hypertelescope images

Aims. We study the possibility of deconvolving hypertelescope images and propose a procedure that can be used provided that the densification factor is small enough to make the process reversible. Methods. We present the simulation of hypertelescope images for an array of cophased densified apertures. We distinguish between two types of aperture densification, one called FAD (full aperture densification) corresponding to Labeyrie’s original technique, and the other FSD (full spectrum densification) corresponding to a densification factor twice as low. Images are compared to the Fizeau mode. A single image of the observed object is obtained in the hypertelescope modes, while in the Fizeau mode the response produces an ensemble of replicas of the object. Simulations are performed for noiseless images and in a photodetection regime. Assuming first that the point spread function (PSF) does not change much over the object extent, we use two classical techniques to deconvolve the images, namely the Richardson-Lucy and image space reconstruction algorithms. Results. Both algorithms fail to achieve satisfying results. We interpret this as meaning that it is inappropriate to deconvolve a relation that is not a convolution, even if the variation in the PSF is very small across the object extent. We propose instead the application of a redilution to the densified image prior to its deconvolution, i.e. to recover an image similar to the Fizeau observation. This inverse operation is possible only when the rate of densification is no more than in the FSD case. This being done, the deconvolution algorithms become efficient. The deconvolution brings together the replicas into a single high-quality image of the object. This is heuristically explained as an inpainting of the Fourier plane. This procedure makes it possible to obtain improved images while retaining the benefits of hypertelescopes for image acquisition consisting of detectors with a small number of pixels.

A Novel Algorithm of Fractal-Wavelet Image Denosing

Image denosing is the first preprocessing step in dealing with image processing where the overall system quality should be improved. So it is a key issue in all image processing researches. Over the past years, fractal-wavelet transforms were introduced in an effort to reduce the blockiness and computational complexity that are inherent in fractal image compression. The essence of fractal image denosing is to predict fractal code of a noiseless image from its noisy observation. From the predicted fractal code, we can generate an estimate of the original image. In the paper, we show how well fractal-wavelet denosing predicts parent wavelet subtrees of the noiseless image. The performance of various fractal-wavelet denosing schemes is compared to that of some standard wavelet thresholding methods. From the several of experimental results, these fractal-based image denosing methods are quite competitive with standard wavelet thresholding methods for image denosing.

A New Fast Iterative Blind Deconvolution Algorithm

Successful blind image deconvolution algorithms require the exact estimation of the Point Spread Function size, PSF. In the absence of any priori information about the imagery system and the true image, this estimation is normally done by trial and error experimentation, until an acceptable restored image quality is obtained. This paper, presents an exact estimation of the PSF size, which yields the optimum restored image quality for both noisy and noiseless images. It is based on evaluating the detail energy of the wave packet decomposition of the blurred image. The minimum detail energies occur at the optimum PSF size. Having accurately estimated the PSF, the paper also proposes a fast double updating algorithm for improving the quality of the restored image. This is achieved by the least squares minimization of a system of linear equations that minimizes some error functions derived from the blurred image. Moreover, a technique is also proposed to improve the sharpness of the deconvolved images, by constrained maximization of some of the detail wavelet packet energies. Simulation results of several examples have verified that the proposed technique manages to yield a sharper image with higher PSNR than classical approaches.

Model-independent Characterization of Charge Diffusion in Thick Fully Depleted CCDs

ABSTRACTWe present a new method to measure charge diffusion in charge-coupled devices (CCDs). The method is based on a statistical characterization of the shapes of charge clouds produced by low-energy X-rays using known properties of the two-dimensional Gaussian point-spread function (PSF). The algorithm produces reliable upper and lower bounds on the size of the PSF for photons converting near the entrance window of a device. It is optimally suited to the case of thick back-illuminated CCDs where the X-ray absorption length is smaller than the silicon thickness and the diffusion scale is of the same order as the pixel size. The only assumptions are that the charge cloud width is a monotonically increasing function of distance from the conversion point to the buried channel, and that the conversion probability is peaked at the surface. Otherwise, no physical models of carrier transport or knowledge of the electric field profile in the CCD are needed. In suboptimal conditions, the upper bound increases and the lower bound is unaffected, so confidence in the correctness of results is retained. The new method has been benchmarked against Monte Carlo simulations and tested on X-ray images measured on thick high-resistivity prototype CCDs for the Large Synoptic Survey Telescope. In Monte Carlo simulations of noiseless images having the optimal diffusion scale, the upper bound approximated the true PSF within 5%, increasing to 10% in simulations with added noise. Even with severely undersampled or truncated PSFs, the method brackets the true value to within 25%. Our method is accurate and computationally efficient and offers a fast and simple experimental setup.

A method for estimating noise variance of CT image

Rank et al. have proposed an algorithm for estimating image noise variance composed of the following three steps: the noisy image is first filtered by a difference operator; a histogram of local signal variances is then computed; and, finally the noise variance is estimated from a statistical evaluation of the histogram. We have verified the accuracy of this algorithm on a CT image by indirect methods, and have shown that this method is able to estimate CT image noise variance with reasonable accuracy, regardless of whether or not the noiseless image is uniform. Further, we have proposed a simple alternative method for the last two steps of the Rank et al. method. However, one must pay attention to the fact that the estimated noise variance will be biased when the nearest two pixels are correlated and that this algorithm does not work well if the assumption of stationarity of noise components is violated.

SAR image processing using adaptive stack filter

Stack filters are a special case of non-linear filters. They have a good performance for filtering images with different types of noise while preserving edges and details. A stack filter decomposes an input image into several binary images according to a set of thresholds. Each binary image is filtered by a Boolean function. The Boolean function that characterizes an adaptive stack filter is optimal and is computed from a pair of images consisting of an ideal noiseless image and its noisy version. In this work the behavior of adaptive stack filters on synthetic aperture radar (SAR) data is evaluated. With this aim, the equivalent number of looks for stack filtered data are calculated to assess the speckle noise reduction capability of this filter. Then a classification of simulated and real SAR images is carried out on data filtered with a stack filter trained with selected samples. The results of a maximum likelihood classification of these data are evaluated and compared with the results of classifying images previously filtered using the Lee and the Frost filters.

Image Edge Respecting Denoising with Edge Denoising by a Designer Nonisotropic Structure Tensor Method

AbstractWe consider image denoising as the problem of removing spurious oscillations due to noise while preserving edges in the images. We will suggest here how to directly make infinitesimal adjustment to standard variational methods of image denoising, to enhance desirable target assumption of the noiseless image. The standard regularization method is used to define a suitable energy functional to penalize the data fidelity and the smoothness of the solution. This energy functional is tailored so that the region with small gradient is isotropically smoothed whereas in a neighborhood of an edge presented by a large gradient smoothing is allowed only along the edge contour. The regularized solution that arises in this fashion is then the solution of a variational principle. To this end the associated Euler — Lagrange equation needs to be solved numerically and the half-quadratic minimization is generally used to linearize the equation and to derive an iterative scheme. We describe here a method to modify Euler — Largrange equation from commonly used energy functionals, in a way to enhance certain desirable preconceived assumptions of the image, such as edge preservation. From an algorithmic point of view, we may deem this algorithm as a smoothing by a local average with an adaptive gradient-based weight. However, this algorithm may result in noisy edges although the edge is preserved and noise is suppressed in the low-gradient regions of the image. The main focus here is to present an edge-preserving regularization in the aforementioned view point, and to provide an alternative and simple way to modify the existing algorithm to mitigate the phenomena of noisy edges without explicitly defining step where we specify an energy functional to be minimized.

The SURE-LET Approach to Image Denoising

We propose a new approach to image denoising, based on the image-domain minimization of an estimate of the mean squared error--Stein's unbiased risk estimate (SURE). Unlike most existing denoising algorithms, using the SURE makes it needless to hypothesize a statistical model for the noiseless image. A key point of our approach is that, although the (nonlinear) processing is performed in a transformed domain--typically, an undecimated discrete wavelet transform, but we also address nonorthonormal transforms--this minimization is performed in the image domain. Indeed, we demonstrate that, when the transform is a "tight" frame (an undecimated wavelet transform using orthonormal filters), separate subband minimization yields substantially worse results. In order for our approach to be viable, we add another principle, that the denoising process can be expressed as a linear combination of elementary denoising processes--linear expansion of thresholds (LET). Armed with the SURE and LET principles, we show that a denoising algorithm merely amounts to solving a linear system of equations which is obviously fast and efficient. Quite remarkably, the very competitive results obtained by performing a simple threshold (image-domain SURE optimized) on the undecimated Haar wavelet coefficients show that the SURE-LET principle has a huge potential.

An enhanced fractal image denoising algorithm

In recent years, there has been a significant development in image denoising using fractal-based method. This paper presents an enhanced fractal predictive denoising algorithm for denoising the images corrupted by an additive white Gaussian noise (AWGN) by using quadratic gray-level function. Meanwhile, a quantization method for the fractal gray-level coefficients of the quadratic function is proposed to strictly guarantee the contractivity requirement of the enhanced fractal coding, and in terms of the quality of the fractal representation measured by PSNR, the enhanced fractal image coding using quadratic gray-level function generally performs better than the standard fractal coding using linear gray-level function. Based on this enhanced fractal coding, the enhanced fractal image denoising is implemented by estimating the fractal gray-level coefficients of the quadratic function of the noiseless image from its noisy observation. Experimental results show that, compared with other standard fractal-based image denoising schemes using linear gray-level function, the enhanced fractal denoising algorithm can improve the quality of the restored image efficiently.

Fractal-wavelet image denoising revisited

The essence of fractal image denoising is to predict the fractal code of a noiseless image from its noisy observation. From the predicted fractal code, one can generate an estimate of the original image. We show how well fractal-wavelet denoising predicts parent wavelet subtress of the noiseless image. The performance of various fractal-wavelet denoising schemes (e.g., fixed partitioning, quadtree partitioning) is compared to that of some standard wavelet thresholding methods. We also examine the use of cycle spinning in fractal-based image denoising for the purpose enhancing the denoised estimates. Our experimental results show that these fractal-based image denoising methods are quite competitive with standard wavelet thresholding methods for image denoising. Finally, we compare the performance of the pixel- and wavelet-based fractal denoising schemes.

A noiseless kilohertz frame rate imaging detector based on microchannel plates read out with the Medipix2 CMOS pixel chip

A new hybrid optical imaging detector is described that is being developed for the next generation adaptive optics (AO) wavefront sensors (WFS) for ground-based telescopes. The detector consists of a photocathode and proximity focused microchannel plates (MCPs) read out by the Medipix2 CMOS pixel ASIC. Each pixel of the Medipix2 device measures 55 × 55 μ m 2 and comprises pre-amplifier, a window discriminator and a 14-bit counter. The 256 × 256 Medipix2 array can be read out noiselessly in 287 μ s . The readout can be electronically shuttered down to a temporal window of a few μ s . The Medipix2 is buttable on three sides to produce 512 × ( n * 256 ) pixel devices. Measurements with ultraviolet light yield a spatial resolution of the detector at the Nyquist limit. Sub-pixel resolution can be achieved using centroiding algorithms. For the AO application, very high continuous frame rates of the order of 1 kHz are required for a matrix of 512 × 512 pixels. The design concepts of a parallel readout board are presented that will allow this fast data throughput. The development status of the optical WFS tube is also explained.

体幹部における呼吸停止下での３検出器型高速回転ＳＰＥＣＴ撮像法の開発―｀９９ｍ´Ｔｃ‐ＭＡＡの臨床応用―

In traditional pulmonary perfusion single photon emission computed tomography (SPECT), respiratory lung motion and cyclically varying changes in lung volume during image acquisition inherently degrade the image sharpness of ill-defined perfusion defects. However, because of the lack of an adequate fast imaging technique, perfusion SPECT has never been acquired under breathhold conditions, whereas breathhold images are commonly used for pulmonary magnetic resonance (MR) and computed tomographic (CT) images. Although a high-speed imaging technique combined with a multidetector SPECT system may enable SPECT images to be obtained during a short period of breathholding, image quality would be degraded owing to decreased radioactivity counts and increased statistical noise. To resolve this problem, we developed an innovative SPECT imaging technique using a triple-head SPECT system and the high-speed-detector rotation-multiplied projection (HSRMP) technique, where a single SPECT image was reconstructed from multiple respiratory dimensional breathhold projection data obtained at the same angle. HSRMP provided noiseless high-quality perfusion SPECT images by compensating for decreased radioactivity counts caused by high-speed imaging, and significantly improved image quality and perfusion defect clarity compared with traditional non-breathhold SPECT images.

Sub-pixel image registration with a maximum likelihood estimator

We present a new method based on a maximum likelihood (ML) estimation of the sub-pixel shift between images of a given object observed with a single instrument. We first study the case of two noisy images and give the ML approach of the registration problem. By means of simulations, we show the gain obtained with this ML solution compared to a classical registration method with an academic noise model (stationary white Gaussian), and then demonstrate the relevance of this ML estimation with a more realistic noise model. We then address the problem of a sequence of low signal frames of the same object. We develop a joint ML approach in which we simultaneously estimate the reference (i.e. the noiseless) image and the shift parameters. The registration accuracy is increased at low photon levels as the number of frames grows, reaching the sub-pixel domain at very low SNR (about 1), when considering 100 frames. When applied to experimental data (thermal IR images of a faint galaxy), both ML methods show their efficiency to recover the resolution in averaged frames and totally outperform the classical cross-correlation.

WE‐A‐I‐617‐01: PET/CT Attenuation Correction and Image Fusion

PET/CT scanners provide intrinsically co‐registered CT and PET images in a single scan session, which can then be viewed together as a fused image showing both anatomy and function. This in particular allows for excellent localization of features in the PET image. Furthermore, with the CT scan performed prior to the PET scan, routine attenuation correction of the acquired PET data is possible. The use of CT images for this purpose in PET/CT scanners has the advantages of shorter scan times and effectively noiseless transmission images as compared to the transmission scans that are used in standalone PET cameras. This does however require the non‐trivial step of transforming the CT images to 511 keV attenuation images that are reprojected to obtain the desired attenuation correction factors. This lecture will describe the use of CT images for attenuation correction in PET/CT scanners and address the advantages and challenges of this procedure and the use of fused images in the clinical environment. Improved attenuation correction procedures to take account of different kVp settings in the CT protocol, as well as the particular challenges in emerging applications such as cardiac PET/CT will be addressed.Educational Objectives:1. Review the principles of attenuation correction in PET and the implementation of this procedure using CT images in PET/CT imaging.2. Understand the particular challenges of CT‐based attenuation correction, including the choice of respiration protocols and other problems of patient motion, the use of CT contrast agent, and the presence of foreign objects in the body.3. Review the advantages of image fusion in PET/CT, and the clinical challenges of registering the CT and PET imaging modalities that remain, such as respiration.

Edge detection and noise removal by use of a partial differential equation with automatic selection of parameters

This work deals with noise removal by the use of an edge preserving method whose parameters are automatically estimated, for any application, by simply providing information about the standard deviation noise level we wish to eliminate. The desired noiseless image u(x), in a Partial Differential Equation based model, can be viewed as the solution of an evolutionary differential equation ut(x) = F(uxx, ux, u, x, t) which means that the true solution will be reached when t ® ¥. In practical applications we should stop the time ''t'' at some moment during this evolutionary process. This work presents a sufficient condition, related to time t and to the standard deviation s of the noise we desire to remove, which gives a constant T such that u(x, T) is a good approximation of u(x). The approach here focused on edge preservation during the noise elimination process as its main characteristic. The balance between edge points and interior points is carried out by a function g which depends on the initial noisy image u(x, t0), the standard deviation of the noise we want to eliminate and a constant k. The k parameter estimation is also presented in this work therefore making, the proposed model automatic. The model's feasibility and the choice of the optimal time scale is evident through out the various experimental results.