Noiseless Image Research Articles

Super-resolution Fourier transforms by optimisation, and ISAR imaging

5A method for super-resolution ISAR imaging, denoising, and error estimation is developed using a novel Fourier transform that exploits the a priori information that the image is sparse, i.e. contains relatively few bright points. The method applies nonlinear optimisation to the complex-valued pixels to estimate the image by minimising its l/sup 1/-norm. Noiseless images require linear programming, while quadratic programming with logarithmic barriers is necessary when complex-valued Gaussian noise is present. The novel Fourier transform, which is referred to as the l/sup 1/-FFT, works with 'missing' data points, making 'jackknife' estimates of the mean and variance of each pixel value possible. These estimates should aid in image classification. This work extends earlier work of Chen, Donoho and Saunders on basis pursuit and denoising to complex signals, by formulating and solving the corresponding complex-valued nonlinear optimisation problems

Harmonic wavelet joint transform correlator: analysis, algorithm, and application to image denoising

We propose analysis and algorithm implementation for a novel harmonic wavelet joint transform correlator (HWJTC) for image noise reduction and automatic target recognition (ATR) applications. We investigate useful properties of the HWJTC and apply our novel correlator in ATR for both noiseless and noisy synthetic aperture radar (SAR) images. Performance is demonstrated in reducing both simulated speckle noise and realistic clutter noise for SAR images. We also compare its performance with that of a well-established image denoising technique such as wavelet packet (WP) processing. The test images used in our simulation are obtained from a CAD simulation model such as Xpatch and Defense Advanced Research Projects Agency (DARPA) Moving and Stationary Target Recognition (MSTAR) program's synthetic aperture radar (SAR) image databases respectively. Simulation results show improved noise tolerance for HWJTC.

An image change detection algorithm based on Markov random field models

This paper addresses the problem of image change detection (ICD) based on Markov random field (MRF) models. MRF has long been recognized as an accurate model to describe a variety of image characteristics. Here, we use the MRF to model both noiseless images obtained from the actual scene and change images (CIs), the sites of which indicate changes between a pair of observed images. The optimum ICD algorithm under the maximum a posteriori (MAP) criterion is developed under this model. Examples are presented for illustration and performance evaluation.

New maximum likelihood motion estimation schemes for noisy ultrasound images

When performing block-matching based motion estimation with the ML estimator, one would try to match blocks from the two images, within a predefined search area. The estimated motion vector is that which maximizes a likelihood function, formulated according to the image formation model. Two new maximum likelihood motion estimation schemes for ultrasound images are presented. The new likelihood functions are based on the assumption that both images are contaminated by a Rayleigh distributed multiplicative noise. The new approach enables motion estimation in cases where a noiseless reference image is not available. Experimental results show a motion estimation improvement with regards to other known ML estimation methods.

Medical image segmentation using a contextual-constraint-based Hopfield neural cube

Neural-network-based image techniques such as the Hopfield neural networks have been proposed as an alternative approach for image segmentation and have demonstrated benefits over traditional algorithms. However, due to its architecture limitation, image segmentation using traditional Hopfield neural networks results in the same function as thresholding of image histograms. With this technique high-level contextual information cannot be incorporated into the segmentation procedure. As a result, although the traditional Hopfield neural network was capable of segmenting noiseless images, it lacks the capability of noise robustness. In this paper, an innovative Hopfield neural network, called contextual-constraint-based Hopfield neural cube (CCBHNC) is proposed for image segmentation. The CCBHNC uses a three-dimensional architecture with pixel classification implemented on its third dimension. With the three-dimensional architecture, the network is capable of taking into account each pixel's feature and its surrounding contextual information. Besides the network architecture, the CCBHNC also differs from the original Hopfield neural network in that a competitive winner-take-all mechanism is imposed in the evolution of the network. The winner-take-all mechanism adeptly precludes the necessity of determining the values for the weighting factors for the hard constraints in the energy function in maintaining feasible results. The proposed CCBHNC approach for image segmentation has been compared with two existing methods. The simulation results indicate that CCBHNC can produce more continuous, and smoother images in comparison with the other methods.

Phase conjugation, imaging in biological media and characterization of quantum noise by parametric image amplification

We review recent results about parametric image amplification obtained in our group and in others. First, forward phase conjugation has been used to restore images through aberrant media and to image objects embedded through thick biological tissues. Second, quantum noise in image amplification has been actively studied by several groups, leading to theoretical demonstration of image entanglement and squeezing, as well as experiments on noiseless image amplification and characterization of the correlations in the spatial fluctuations of amplified images.

Estimating voxel volume fractions of trabecular bone on the basis of magnetic resonance images acquiredin vivo

A method is described for extracting information from images acquired in the limited spatial resolution regime in which the structures to be identified are smaller than the image voxel size. Under these conditions and in the presence of noise, the voxel intensity histogram is monomodal; therefore, segmenting the image with an intensity threshold is inaccurate. The present method, which is applicable to two-phase materials of discrete intensity, relies on iterative deconvolution to obtain a noiseless histogram. A noiseless image is then generated on the basis of the noiseless histogram and the unprocessed image. In the current work, the method is referred to as bone volume fraction (BVF) mapping, since it has been applied to determine the spatial distribution of trabecular bone. BVF computed from maps generated on the basis of in vivo magnetic resonance images of the human radius has been shown to compare well with bone mineral density (R2 = 0.8). The accuracy and precision of BVF measurements have been further evaluated by using a BVF map of the radius as a gold standard and then adding different levels of noise to generate test images. The results suggest that the error in BVF is <0.01 for SNR > 8. © 1999 John Wiley & Sons, Inc. Int J Imaging Syst Technol, 10, 186–198, 1999

Multidimensional Fuzzy Filtering using Statistical Parameters

A multidimensional filtering technique is proposed using fuzzy logic ideas and based on two locally estimated statistical parameters. This technique is applied to the 3D space of color images. Two fuzzy sets for each fuzzy variable are considered. These fuzzy variables assign the corresponding part of the image to a specific class. The parameters of the fuzzy sets are derived in respect to the corresponding ’’crisp‘‘ sets attributed to the ideal noiseless image. An additional fuzzy selection process is also introduced by applying a distance dependent weighted average as the filtering action of each rule. Both signal and noise characteristics are taken into account in the filtering process. Experimental results show that these fuzzy non-linear filters work very well.

Optimizing Edge Detectors for Robust Automatic Threshold Selection: Coping with Edge Curvature and Noise

The Robust Automatic Threshold Selection algorithm was introduced as a threshold selection based on a simple image statistic. The statistic is an average of the grey levels of the pixels in an image weighted by the response at each pixel of a specific edge detector. Other authors have suggested that many edge detectors may be used within the context of this method instead. A simple proof of this is given, including an extension to any number of image dimensions, and it is shown that in noiseless images with straight line edges these statistics all yield an optimum threshold. Biases caused by curvature of edges and by noise (uniform Gaussian and Poisson) are explored theoretically and on synthetic 2-D images. It is shown that curvature bias may be avoided by proper selection of the edge detector, and a comparison of two noise bias reduction schemes is given. Criteria for optimizing edge detectors are given and the performances of eight edge detectors are investigated in detail. The best results were obtained using two edge detectors which compute an approximation of the square of the gradient. It is shown that this conclusion can be extended to 3-D. Least sensitivity to noise was obtained when using 3 × 3 Sobel filter kernels to approximate partial derivatives inxandy.

Algorithms for Scanned Probe Microscope Image Simulation, Surface Reconstruction, and Tip Estimation.

To the extent that tips are not perfectly sharp, images produced by scanned probe microscopies (SPM) such as atomic force microscopy and scanning tunneling microscopy are only approximations of the specimen surface. Tip-induced distortions are significant whenever the specimen contains features with aspect ratios comparable to the tip’s. Treatment of the tip-surface interaction as a simple geometrical exclusion allows calculation of many quantities important for SPM dimensional metrology. Algorithms for many of these are provided here, including the following: (1) calculating an image given a specimen and a tip (dilation), (2) reconstructing the specimen surface given its image and the tip (erosion), (3) reconstructing the tip shape from the image of a known “tip characterizer” (erosion again), and (4) estimating the tip shape from an image of an unknown tip characterizer (blind reconstruction). Blind reconstruction, previously demonstrated only for simulated noiseless images, is here extended to images with noise or other experimental artifacts. The main body of the paper serves as a programmer’s and user’s guide. It includes theoretical background for all of the algorithms, detailed discussion of some algorithmic problems of interest to programmers, and practical recommendations for users.

ニューラルネットワークによるポラリメトリックＳＡＲデータの分類

In this paper, we consider classification methods using neural networks for multi-frequency polarimetric SAR data. This data was obtained by SIR-C inside Space Shuttle (Space Radar Lab I) . We propose a learning algorithm for competitive neural networks and find the most effective feature vector. The proposed algorithm is as follows: First, weight vectors are trained by one of three basic algorithms (LVQ1, LVQ2.1, and OLVQ1) . Next, they are re-trained and tuned up by the LVQ2.1 for a narrower window area around the category boundaries.In these experiments, we demonstrate the re-training effects. Then nine types of feature vectors are adopted to the classification methods, which in turn are compared to the conventional methods such as the maximum likelihood method and the layered neural networks (BP) with respect to average accuracy, classification image quality, and computational time. As a result, the proposed method, especially, OLVQ1 + LVQ2.1 outperforms other methods and the feature vector which is composed of back-scattering power of cross polarization for L and C bands generates the best classification accuracies and noiseless images.

Multiresolution edge detection techniques

In this paper, we propose multiresolution edge detection techniques for noiseless images and noisy ones which are contaminated by additive Gaussian noise or multiplicative noise. For edge detection, they determine automatically the resolution of a pixel by using pertinent discontinuity measures. The mode of the discontinuity measure γ calculated over the (2 p + 1) × (2 p + 1) windows is used to determine the resolution of a given pixel and the edge operator with a varying scale is applied to the pixel depending on its resolution. For noiseless images and noisy images contaminated by additive Gaussian noise, the local variance γ a is used as an edgeness measure. For multiplicative speckle images, the statistical quantities such as the ratio γ s of the variance to mean-square, or the maximum difference γ d between the real and theoretical cumulative density functions (CDF's) are employed as discontinuity measures. The application of the proposed pixelwise resolution determination algorithm to the conventional Canny and LoG edge detectors, for example, gives the proposed multiresolution edge detection techniques. Their effectiveness is shown via computer simulations.

ニューラルネットによる画像ノイズ除去の応用 溶接接合部開先形状画像の信号処理

On using the signal processing ability of Backpropagation Model of Neural Network, the noise elimination is performed on the image of a joint part of arc welding. The object image, which is obtained by the slit light slice method, is often used for the detection of joint configurationis on the purpose of welding automation. The noise caused by the welding arc is superimposed on the image and disturbs the detection. Such noise is eliminated through the learning process of Neural Network which is carried out for the noise imposed input images and the noiseless teacher images. The structure of the network, the image input method, the selection of the teacher image and the method to recover the resolution of the processed result are investigated experimentally. It is shown that the noise elimination is successfully done as the result.

Performance of an optimal subset of Zernike features for pattern classification

This paper presents a technique of selecting an optimal number of features from the original set of features. Due to the large number of features considered, it is computationally more efficient to select a subset of features that can discriminate as well as the original set. The subset of features is determined using stepwise discriminant analysis. The results of using such a scheme to classify scaled, rotated, and translated binary images and also images that have been perturbed with random noise are reported. The features used in this study are Zernike moments, which are the mapping of the image onto a set of complex orthogonal polynomials. The performance of using a subset is examined through its comparison to the original set. The classifiers used in this study are neural network and a statistical nearest neighbor classifier. The back-propagation learning algorithm is used in training the neural network. The classifers are trained with some noiseless images and are tested with the remaining data set. When an optimal subset of features is used, the classifers performed almost as well as when trained with the original set of features.

Constructing noise-reducing operators from training images

We discuss constructing non-linear noise reduction operators on binary images using a training set of noiseless images. We extract from the training set a probability distribution over local neighborhoods. Our operator changes pixel values when such a change turns a low probability neighborhood into a high probability one.

Fresnel ping-pong algorithm for two-plane computer-generated hologram display

Image formation of three-dimensional objects suffers from out-of-focus noise if the light is coherent. Nevertheless, it is possible to generate two noiseless image intensities (I(A)and I(B)) at two depth locations by means of a single computer hologram. The phases related to I(A) and I(B) provide design freedom. To accomplish this goal, we use a ping-pong algorithm that bounces back and fourth between the two planes. Our ping-pong algorithm is the fourth member of a family of algorithms. The first member is known as the Gerchberg-Saxton algorithm. Basic considerations and experimental results are presented. Details of the algorithm are explained in Appendix A.

Model-Based Classification of Quadric Surfaces

Model-based 3D object recognition systems using range imagery typically employ entirely data-driven procedures for segmentation and surface classification. However, some recognition environments may contain only objects whose surface types and parameters are known a priori and can therefore be exploited by the early-processing steps used in the recognition system. We propose a new suite of model-driven techniques for identification of quadric surfaces (cones, cylinders, and spheres) in segmented range imagery. The methods employ surface positions and surface normal estimates in combination with the known parameters of surfaces in a database of object models. Second-derivative quantities (i.e., surface curvatures) are not used. The free parameters of cylinders and spheres are accumulated using a Hough transform, and free parameters of cones are estimated using a regression procedure. Experiments are presented for numerous scenes of both real and synthetic objects including part jumbles, objects in many poses, objects containing concave and convex surfaces, and noiseless and noisy synthetic range images of objects. Our experimental results show that the proposed surface classification methods can accurately recover surface parameters from both synthetic and real images, making them viable for environments with partial knowledge of surface type and parameters.

A comparison between Fourier-Mellin descriptors and moment based features for invariant object recognition using neural networks

Two algorithms for invariant object recognition are compared using a neural network and two statistical classifiers as classifiers for features taken from noiseless and noisy images. It was found that the Fourier-Mellin descriptors perform as well as moment based features for noiseless images but perform significantly better when noise is added. Neural networks perform better than statistical classifiers when noise is added. It was also found that Fourier-Mellin descriptors are very sensitive to uncertainty in the position of the object centroid whereas the moment based features provide good performance when the object is not presented at the centroid.

Minimum entropy deconvolution for restoration of blurred two-tone images

A blind image restoration algorithm for noiseless blurred two-tone images based on the criterion of minimum entropy is presented. Performance of the algorithm is demonstrated by simulation.

Segmentation through variable-order surface fitting

The solution of the segmentation problem requires a mechanism for partitioning the image array into low-level entities based on a model of the underlying image structure. A piecewise-smooth surface model for image data that possesses surface coherence properties is used to develop an algorithm that simultaneously segments a large class of images into regions of arbitrary shape and approximates image data with bivariate functions so that it is possible to compute a complete, noiseless image reconstruction based on the extracted functions and regions. Surface curvature sign labeling provides an initial coarse image segmentation, which is refined by an iterative region-growing method based on variable-order surface fitting. Experimental results show the algorithm's performance on six range images and three intensity images. >