Noisy Regions Research Articles

Visual Attention Network for Low-Dose CT

Noise and artifacts are intrinsic to low dose CT (LDCT) data acquisition, and will significantly affect the imaging performance. Perfect noise removal and image restoration is intractable in the context of LDCT due to the statistical and technical uncertainties. In this paper, we apply the generative adversarial network (GAN) framework with a visual attention mechanism to deal with this problem in a data-driven/machine learning fashion. Our main idea is to inject visual attention knowledge into the learning process of GAN to provide a powerful prior of the noise distribution. By doing this, both the generator and discriminator networks are empowered with visual attention information so they will not only pay special attention to noisy regions and surrounding structures but also explicitly assess the local consistency of the recovered regions. Our experiments qualitatively and quantitatively demonstrate the effectiveness of the proposed method with clinic CT images.

An FPGA-based real-time occlusion robust stereo vision system using semi-global matching

Stereo matching approaches are an appealing choice for acquiring depth information in a number of video processing applications. It is desirable that these solutions generate dense, robust disparity maps in real time. However, occlusion regions may disturb the applications that need these maps. Among the best of these approaches is the semi-global matching (SGM) technique. This paper presents an FPGA-based stereo vision system based on SGM. This system calculates disparity maps by streaming, which are scalable to several resolutions and disparity ranges. To increase the robustness of the SGM technique even further, the present work has implemented a combination of the gradient filter and the sampling-insensitive absolute difference in the pre-processing phase. Furthermore, as a post-processing step, this paper proposes a novel streaming architecture to detect noisy and occluded regions. The FPGA-based implementations of the proposed stereo matching system in two distinct heterogeneous architecture (GPP—general purpose processor, and FPGA) were evaluated using the Middlebury stereo vision benchmark. The achieved results reported a frame rate of 25 FPS for the disparity maps processing in HD resolution (1024 $$\times$$ 768 pixels), with 256 disparity levels. The results have demonstrated that the memory utilization, processing performance, and accuracy are among the best of FPGA-based stereo vision systems.

Pairing an arbitrary regressor with an artificial neural network estimating aleatoric uncertainty

We suggest a general approach to quantification of different forms of aleatoric uncertainty in regression tasks performed by artificial neural networks. It is based on the simultaneous training of two neural networks with a joint loss function and a specific hyperparameter λ > 0 that allows for automatically detecting noisy and clean regions in the input space and controlling their relative contribution to the loss and its gradients. After the model has been trained, one of the networks performs predictions and the other quantifies the uncertainty of these predictions by estimating the locally averaged loss of the first one. Unlike in many classical uncertainty quantification methods, we do not assume any a priori knowledge of the ground truth probability distribution, neither do we, in general, maximize the likelihood of a chosen parametric family of distributions. We analyze the learning process and the influence of clean and noisy regions of the input space on the loss surface, depending on λ. In particular, we show that small values of λ increase the relative contribution of clean regions to the loss and its gradients. This explains why choosing small λ allows for better predictions compared with neural networks without uncertainty counterparts and those based on classical likelihood maximization. Finally, we demonstrate that one can naturally form ensembles of pairs of our networks and thus capture both aleatoric and epistemic uncertainty and avoid overfitting.

Two-network Kuramoto-Sakaguchi model under tempered stable Lévy noise.

We examine a model of two interacting populations of phase oscillators labeled "blue" and "red." To this we apply tempered stable Lévy noise, a generalization of Gaussian noise where the heaviness of the tails parametrized by a power law exponent α can be controlled by a tempering parameter λ. This system models competitive dynamics, where each population seeks both internal phase synchronization and a phase advantage with respect to the other population, subject to exogenous stochastic shocks. We study the system from an analytic and numerical point of view to understand how the phase lag values and the shape of the noise distribution can lead to steady or noisy behavior. Comparing the analytic and numerical studies shows that the bulk behavior of the system can be effectively described by dynamics in the presence of tilted ratchet potentials. Generally, changes in α away from the Gaussian noise limit 1<α<2 disrupt the locking between blue and red, while increasing λ acts to restore it. However, we observe that with further decreases of α to small values α≪1, with λ≠0, locking between blue and red may be restored. This is seen analytically in a restoration of metastability through the ratchet mechanism, and numerically in transitions between periodic and noisy regions in a fitness landscape using a measure of noise. This nonmonotonic transition back to an ordered regime is surprising for a linear variation of a parameter such as the power law exponent and provides a mechanism for guiding the collective behavior of such a complex competitive dynamical system.

Overlapped Apple Fruit Yield Estimation using Pixel Classification and Hough Transform

Researchers proposed various visual based methods for estimating the fruit quantity and performing qualitative analysis, they used ariel and ground vehicles to capture the fruit images in orchards. Fruit yield estimation is a challenging task with environmental noise such as illumination changes, color variation, overlapped fruits, cluttered environment, and branches or leaves shading. In this paper, we proposed a learning free fast visual based method to correctly count the apple fruits tightly overlapped in a complex outdoor orchard environment. We first carefully build the color based HS model to perform the color based segmentation. This step extracts the apple fruits from the complex orchard background and produces the blobs representing apples along with the additional noisy regions. We used the fine tuned morphological operators to refine the blobs received from the previous step and remove the noisy regions fol-lowed by the Gaussian smoothing. Finally we treated the circular shaped blobs with Hough Transform algorithm to calculate the center coordinates of each apple edge and the method correctly locates the apples in the images. The results ensures the proposed algorithm successfully detects and count apple fruits in the images captured from apple orchard and outperforms the standard state of the art contoured based method.

An Improved Density-Based Approach to Spatio-Textual Clustering on Social Media

Density-based spatial clustering of applications with noise (DBSCAN) is the most commonly used density-based clustering algorithm but may not be sufficient when the input data type is heterogeneous in terms of textual description. When we aim to discover clusters of geo-tagged records relevant to a particular point of interest (POI) on social media, examining only one type of input data (e.g., the tweets relevant to a POI) may draw an incomplete picture of clusters due to noisy regions. To overcome this problem, we introduce DBSTexC, a newly defined density-based clustering algorithm using spatio-textual information on social media (e.g., Twitter). We first characterize the POI-relevant and POI-irrelevant geotagged tweets as the texts that include and do not include a POI name or its semantically coherent variations, respectively. By leveraging the proportion of the POI-relevant and POI-irrelevant tweets, the proposed algorithm demonstrates much higher clustering performance than the DBSCAN case in terms of F 1 score and its variants. While DBSTexC performs exactly as DBSCAN with the textually homogeneous inputs, it far outperforms DBSCAN with the textually heterogeneous inputs. Furthermore, to further improve the clustering quality by fully capturing the geographic distribution of geo-tagged points, we present fuzzy DBSTexC (F-DBSTexC), an extension of DBSTexC, which incorporates the notion of fuzzy clustering into the DBSTexC. We then demonstrate the consistent superiority of F-DBSTexC over the original DBSTexC via intensive experiments. The computational complexity of our algorithms is also analytically and numerically shown.

Nonlinear diffusion based image segmentation using two fast algorithms

In this paper, a new variational model is proposed for image segmentation based on active contours, nonlinear diffusion and level sets. It includes a Chan-Vese model-based data fitting term and a regularized term that uses the potential functions (PF) of nonlinear diffusion. The former term can segment the image by region partition instead of having to rely on the edge information. The latter term is capable of automatically preserving image edges as well as smoothing noisy regions. To improve computational efficiency, the implementation of the proposed model does not directly solve the high order nonlinear partial differential equations and instead exploit the efficient alternating direction method of multipliers (ADMM), which allows the use of fast Fourier transform (FFT), analytical generalized soft thresholding equation, and projection formula. In particular, we creatively propose a new fast algorithm, normal vector projection method (NVPM), based on alternating optimization method and normal vector projection. Its stability can be the same as ADMM and it has faster convergence ability. Extensive numerical experiments on grey and colour images validate the effectiveness of the proposed model and the efficiency of the algorithms.

Two-Step Modified Nash Equilibrium Method for Medical Image Segmentation Based on an Improved C-V Model

One of the most established region-based segmentation methods is the region based C-V model. This method formulates the image segmentation problem as a level set or improved level set clustering problem. However, the existing level set C-V model fails to perform well in the presence of noisy and incomplete data or when there is similarity between the objects and background, especially for clustering or segmentation tasks in medical images where objects appear vague and poorly contrasted in greyscale. In this paper, we modify the level set C-V model using a two-step modified Nash equilibrium approach. Firstly, a standard deviation using an entropy payoff approach is employed and secondly a two-step similarity clustering based approach is applied to the modified Nash equilibrium. One represents a maximum similarity within the clustered regions and the other the minimum similarity between the clusters. Finally, an improved C-V model based on a two-step modified Nash equilibrium is proposed to smooth the object contour during the image segmentation. Experiments demonstrate that the proposed method has good performance for segmenting noisy and poorly contrasting regions within medical images.

HAHap: a read-based haplotyping method using hierarchical assembly.

BackgroundThe need for read-based phasing arises with advances in sequencing technologies. The minimum error correction (MEC) approach is the primary trend to resolve haplotypes by reducing conflicts in a single nucleotide polymorphism-fragment matrix. However, it is frequently observed that the solution with the optimal MEC might not be the real haplotypes, due to the fact that MEC methods consider all positions together and sometimes the conflicts in noisy regions might mislead the selection of corrections. To tackle this problem, we present a hierarchical assembly-based method designed to progressively resolve local conflicts.ResultsThis study presents HAHap, a new phasing algorithm based on hierarchical assembly. HAHap leverages high-confident variant pairs to build haplotypes progressively. The phasing results by HAHap on both real and simulated data, compared to other MEC-based methods, revealed better phasing error rates for constructing haplotypes using short reads from whole-genome sequencing. We compared the number of error corrections (ECs) on real data with other methods, and it reveals the ability of HAHap to predict haplotypes with a lower number of ECs. We also used simulated data to investigate the behavior of HAHap under different sequencing conditions, highlighting the applicability of HAHap in certain situations.

New FCM Segmentation Approach Based on Multi-Resolution Analysis

This article presents a modified Fuzzy C Means segmentation approach based on multi-resolution image analysis. Fuzzy C-Means standard methods are improved through fuzzy clustering at different image resolution levels by propagating fuzzy membership values pyramidally from a lower to a higher level. Processing at a lower resolution image level provides a rough pixel classification result, thus, a pixel is assigned to a cluster to which the majority of its neighborhood pixels belongs. The aim of fuzzy clustering with multi-resolution images is to avoid pixel misclassification according to the spatial cluster of the neighbourhood of each pixel in order to have more homogeneous regions and eliminate noisy regions present in the image. This method is tested particularly on samples and medical images with gaussian noise by varying multiresolution parameter values for better analysis. The results obtained after multi-resolution clustering are giving satisfactory results by comparing this approach with standard FCM and spatial FCM ones.

Region-Aware Image Denoising by Exploring Parameter Preference

The goal of an image denoising algorithm is to preserve the details of clean images while reducing the noise in noisy images. Some existing image denoising algorithms preserve the details by using external information. However, external information needs to be obtained from external images or regions similar to the noisy images or regions. In this letter, we propose a region-aware image denoising algorithm (RAID) by exploring parameter preference. The proposed RAID algorithm is based on the observation that different regions of noisy images prefer denoised results that differ due to being obtained with different denoising parameters. The RAID algorithm, first, measures the extent of preferences of denoising parameters for different image regions. Then, it combines the denoised results obtained by using the various denoising parameters according to the extent of preferences determined in the previous step to get the final denoised image. Experimental results show that the proposed RAID algorithm can be combined with some existing image denoising algorithms to improve their denoising performance. Our denoised results are better at preserving the details while reducing the noise than existing algorithms which use the same parameter for the whole image.

FIRE-DES++: Enhanced online pruning of base classifiers for dynamic ensemble selection

Dynamic Ensemble Selection (DES) techniques aim to select one or more competent classifiers for the classification of each new test sample. Most DES techniques estimate the competence of classifiers using a given criterion over the region of competence of the test sample, usually defined as the set of nearest neighbors of the test sample in the validation set. Despite being very effective in several classification tasks, DES techniques can select classifiers that classify all samples in the region of competence as being from the same class. The Frienemy Indecision REgion DES (FIRE-DES) tackles this problem by pre-selecting classifiers that correctly classify at least one pair of samples from different classes in the region of competence of the test sample. However, FIRE-DES applies the pre-selection for the classification of a test sample if and only if its region of competence is composed of samples from different classes (indecision region), even though this criterion is not reliable for determining if a test sample is located close to the borders of classes (true indecision region) when the region of competence is obtained using classical nearest neighbors approach. Because of that, FIRE-DES mistakes noisy regions for true indecision regions, leading to the pre-selection of incompetent classifiers, and mistakes true indecision regions for safe regions, leaving samples in such regions without any pre-selection. To tackle these issues, we propose the FIRE-DES++, an enhanced FIRE-DES that removes noise and reduces the overlap of classes in the validation set; and defines the region of competence using an equal number of samples of each class, avoiding selecting a region of competence with samples of a single class. Experiments are conducted using FIRE-DES++ with 8 different dynamic selection techniques on 64 classification datasets. Experimental results show that FIRE-DES++ increases the classification performance of all DES techniques considered in this work, outperforming FIRE-DES with 7 out of the 8 DES techniques, and outperforming state-of-the-art DES frameworks.

Real-time shadow detection using multi-channel binarization and noise removal

High-quality automatic shadow detection remains a challenging problem in image processing and computer vision. Existing techniques for shadow detection typically make use of deep learning strategies to obtain accurate shadow detection results, at the cost of demanding high processing time, making their use unsuitable for augmented reality and robotic applications. In this paper, we propose a novel approach to perform high-quality shadow detection in real time. To do so, we convert an input image into different color spaces to perform multi-channel binarization and detect different shadow regions in the image. Then, a filtering algorithm is proposed to remove the noisy false-positive shadow regions on the basis of their sizes. Experimental results evaluated in two different datasets show that the proposed approach may run entirely on the GPU, requiring only $$\approx$$ 13 ms to detect shadows in an image with $$3840 \times 2160$$ (4k) resolution. That makes our approach about 1.8 (66$$\times$$) to 4.6 (37,284$$\times$$) orders of magnitude faster than related work for 4k resolution images, at the cost of only $$\approx$$ 5% of accuracy loss compared to the best results achieved for each dataset.

Change Detection in Heterogenous Remote Sensing Images via Homogeneous Pixel Transformation.

The change detection in heterogeneous remote sensing images remains an important and open problem for damage assessment. We propose a new change detection method for heterogeneous images (i.e., SAR and optical images) based on homogeneous pixel transformation (HPT). HPT transfers one image from its original feature space (e.g., gray space) to another space (e.g., spectral space) in pixel-level to make the pre-event and post-event images represented in a common space for the convenience of change detection. HPT consists of two operations, i.e., the forward transformation and the backward transformation. In forward transformation, for each pixel of pre-event image in the first feature space, we will estimate its mapping pixel in the second space corresponding to post-event image based on the known unchanged pixels. A multi-value estimation method with noise tolerance is introduced to determine the mapping pixel using -nearest neighbors technique. Once the mapping pixels of pre-event image are available, the difference values between the mapping image and the post-event image can be directly calculated. After that, we will similarly do the backward transformation to associate the post-event image with the first space, and one more difference value for each pixel will be obtained. Then, the two difference values are combined to improve the robustness of detection with respect to the noise and heterogeneousness (modality difference) of images. Fuzzy-c means clustering algorithm is employed to divide the integrated difference values into two clusters: changed pixels and unchanged pixels. This detection results may contain some noisy regions (i.e., small error detections), and we develop a spatial-neighbor-based noise filter to further reduce the false alarms and missing detections using belief functions theory. The experiments for change detection with real images (e.g., SPOT, ERS, and NDVI) during a flood in U.K. are given to validate the effectiveness of the proposed method.

A Heuristic Approach to Detect and Localize Text on Arabic News Video

Automatic text detection in video sequences remains a challenging problem due to the variety of sizes, colors and the presence of complex background. In this paper, we attempt to solve this problem by proposing a robust detection–validation schema for text localization in Arabic news video. Candidate text regions are first detected by using a hybrid method which combines MSER detector and edge information. Then, these regions are grouped using morphological operators. Finally, a verification process is applied to remove noisy non-text regions including specific features for Arabic text. Performance and efficacy of the proposed text detection approach have been tested By using Arabic-Text-in-Video database (AcTiV-DB).

Synthetic aperture radar image change detection based on frequency-domain analysis and random multigraphs

With the development of earth observation programs, many multitemporal synthetic aperture radar (SAR) images over the same geographical area are available. It is demanding to develop automatic change detection techniques to take advantage of these images. Most existing techniques directly analyze the difference image (DI), and therefore, they are easily affected by the speckle noise. We proposed an SAR image change detection method based on frequency-domain analysis and random multigraphs. The proposed method follows a coarse-to-fine procedure: in the coarse changed regions localization stage, frequency-domain analysis is utilized to select distinctive and salient regions from the DI. Therefore, nonsalient regions are neglected, and noisy unchanged regions incurred by the speckle noise are suppressed. In the fine changed regions classification stage, random multigraphs are employed as the classification model. By selecting a subset of neighborhood features to create graphs, the proposed method can efficiently exploit the nonlinear relations between multitemporal SAR images. The experimental results on two real SAR datasets and one simulated dataset have demonstrated the effectiveness of the proposed method.

Steganography using Gabor filter and anisotropic diffusion

In order to improve the statistical security and the ability of resistance to steganalysis of adaptive steganography, an adaptive steganography algorithm based on Gabor filters and anisotropic diffusion is proposed, which can be utilized in an arbitrary domain. Compared with other adaptive embedding schemes, this method can define image distortion more accurately. Firstly, Gabor filters with multi-orientation and multi-scale is employed to determine the noisy regions or textures of cover image. Then the P-M anisotropic diffusion model is utilized to make image elements with low distortion cost relatively cluster in complex areas of image, and better retain the image edge information. At last, the syndrome-trellis codes is used to realize the actual embedding method to minimize the expected distortion for a given payload. The experimental results illustrate that the proposed method can significantly improve the accuracy of embedding in noisy regions and textures, and gains a better performance on resisting the current state-of-the-art steganalysis over prior algorithms.

Anisotropic complex diffusion filtering for comparison of the vortex identification methods in homogeneous turbulence

The vortex identification methods are mutually compared on the basis of extracting coherent vortices and isolating incoherent regions from a filtered turbulent velocity field. Those identification schemes are the high rotation Q−criterion, the second large eigen value λ2−definition, the Δ−method, the vorticity magnitude |ω| and the new omega identification method ωnew. The velocity field of a forced homogeneous isotropic turbulent flow is obtained by simulating the lattice Boltzmann method (LBM) in a periodical box with resolutions of 1283 and 2563, respectively. The anisotropic diffusion model (ADM) is used to filter the velocity field into coherent and incoherent parts and finally the four identification methods are applied to identify the organized vortices as well as the noisy regions. The probability distribution functions (PDFs), the joint PDFs (JPDFs) and the spectrum of the extracted fields for the different identification methods are examined in details. Results show that the JPDFs have different shapes in the coherent cases, for example, the (λ2−Δ) and (Q−Δ)-JPDFs for the non-filtered and coherent fields show an ellipse-like shape where the large axis is found in the Δ direction. The (|ω|−Δ)-JPDFs for the non-filtered and coherent fields show a half circle - like shape. The (|ω|−Q)-JPDFs for the non - filtered and coherent fields inclined towards the positive direction of Q and it is longer at higher values of |ω|. The (|ω|−λ2)-JPDFs for the non-filtered and coherent fields inclined towards the negative values of λ2 and it is also longer for high values of |ω|. The incoherent JPDFs for the Δ method with other identification methods are found constant and their maximum existence are mostly found for moderate values of Q and λ2 at different values of Δ. However, the incoherent JPDFs (|ω|−Q) and (|ω|−λ2) are almost similar and the incoherent JPDF for (λ2−Q) is different, where most overlapping regions are found at negative values of Q and positive values of λ2. There is a tendency to have a universal JPDFs between the different identification methods where the same JPDF shapes can be observed in the 1283 and 2563 cases.

Tissue characterization with depth-resolved attenuation coefficient and backscatter term in intravascular optical coherence tomography images.

An important application of intravascular optical coherence tomography (IVOCT) for atherosclerotic tissue analysis is using it to estimate attenuation and backscatter coefficients. This work aims at exploring the potential of the attenuation coefficient, a proposed backscatter term, and image intensities in distinguishing different atherosclerotic tissue types with a robust implementation of depth-resolved (DR) approach. Therefore, the DR model is introduced to estimate the attenuation coefficient and further extended to estimate the backscatter-related term in IVOCT images, such that values can be estimated per pixel without predefining any delineation for the estimation. In order to exclude noisy regions with a weak signal, an automated algorithm is implemented to determine the cut-off border in IVOCT images. The attenuation coefficient, backscatter term, and the image intensity are further analyzed in regions of interest, which have been delineated referring to their pathology counterparts. Local statistical values were reported and their distributions were further compared with a two-sample t-test to evaluate the potential for distinguishing six types of tissues. Results show that the IVOCT intensity, DR attenuation coefficient, and backscatter term extracted with the reported implementation are complementary to each other on characterizing six tissue types: mixed, calcification, fibrous, lipid-rich, macrophages, and necrotic core.

Real‐time 3D ultrasound denoising based on adaptive regularisation Savitzky–Golay filter

The demand of the fast acquisition in 3D ultrasound (3D-US) imaging leads to the dense population of speckle noise which causes the difficulty in medical diagnosis. In this Letter, 3Dadaptive regularisation Savitzky–Golay (3D ARSG) filter is proposed as the real-time filter for removing speckle noise in 3D-US imaging. The 3D ARSG filter uses the local intensity homogeneity to classify between noisy and texture regions. The intensity in the noisy regions is heavily smoothed, while the resolvable edges remain sharp. The experiment on the denoising of the 3D simulated models and the 3D-US images indicated that 3D ARSG filter provided the image with far less noise and crisper objects’ boundary than conventional methods. In addition, the proposed filter preserves the structural details with regard to the dimension and can denoise images in real time.