Edge-preserving Smoothing Filter Research Articles

Single-shot enhanced imaging through multimode fiber based on self-learning networks

High-fidelity imaging through a multimode fiber (MMF) plays a crucial role in various fields such as medicine, communication, and detection. However, the optical transmission matrix of the MMF suffers from dimensionality reduction. This case results in poor reconstruction quality such as low resolution and noise. In this paper, an unsupervised self-learning circulate learning network is employed to enhance a single degraded image without ground truth images. Moreover, an edge-preserving smoothing filter is applied to address the heavy noise problem of the reconstructed images. Experimental results demonstrate that the proposed method can improve the dimensionality and fidelity of the reconstructed target. Compared to traditional transmission matrix-based reconstruction methods, we have a competitive advantage in terms of evaluation metrics. The proposed method further advances the development of imaging through a multimode fiber.

Hyperspectral Image Classification Using Improved Active Deep Learning Framework Including Edge Preserving Filters

To extract valuable information from satellite data for applications such as agriculture, geological research, and environmental monitoring, the classification of hyperspectral images is an essential task. Labeling each pixel in this process is time-consuming and requires financial resources. To this end, working with a small number of samples is very important. In order to provide high classification performances with a limited number of samples, this paper aims to enhance the performance with an active learning framework. The framework incorporates dimensionality reduction, an edge-preserving filter, and active learning steps. From this perspective, we investigated different edge-preserving filter methods to analyze the effects on performance. By combining edge-preserving filters with dimensionality reduction, the study presents a unique method that improves classification performance while maintaining image quality and reducing noise. The following five edge-preserving smoothing filters are evaluated: weighted least squares (WLS), Joint-Histogram weighted median filter (Joint WMF), fast global image smoother (FGS), bilateral filter (BF), and static/dynamic (SD). Our experiments demonstrate that compared to the reference research (CNN+AL+MRF), the proposed framework increased overall and average accuracies about 2-5% for Indian Pines, Pavia University, and Salinas datasets.

Edge-preserving smoothing filter using fast M-estimation method with an automatic determination algorithm for basic width

Digital filtering is essential for digital imaging, image recognition, and super-resolution technology. For example, the presence of noise in images captured by digital cameras causes deterioration of the image quality and image recognition rate. In order to improve the image recognition rate, noise reduction and edge preservation must be performed during preprocessing. Noise is generally reduced using low-pass filters, such as the Gaussian filter. Although they reduce noise, such filters also have the properties of blurring edge. A strong edge blur reduces the accuracy of the feature detection in image recognition. Therefore, in our previous study, a fast M-estimation Gaussian filter for images (FMGFI) was proposed as an image filter that simultaneously achieves denoising and edge preservation. In the FMGFI, the setting of the optimal basic width of the 2nd order B-spline basis functions is important for achieving simultaneous denoising and edge preservation. In this method, the optimal basic width of the FMGFI was determined not only by manually setting the basic width but also by human judgment of the filtered images. Consequently, the inability to automatically determine the optimal basic width hindered efficient denoising during image processing Therefore, in this research, we develop and propose a method that can automatically determine the optimal basic width of the FMGFI. The previously proposed method calculates using the same basic width for all the pixels over the entire image; in contrast, the proposed method calculates using the basic width automatically determined for each pixel. The experiments confirmed that the method proposed in this study achieves higher denoising and edge preservation performance than the ones used in previous research. The results also showed that it has the highest denoising performance against salt-and-pepper noise as compared to other filters: non-local mean filter, Gaussian filter, median filter, bilateral filter, adaptive bilateral filter, and FMGFI. The experimental results for the Gaussian noise sowed that the proposed method has the same denoising and edge preservation performance as the other filters in visual evaluation. From the above, the proposed method is expected to contribute to efficient denoising and improvement of image quality by using it as a preprocessing.

GRF: guided residual fusion for pansharpening

ABSTRACT Fusion of a multispectral (MS) image and a panchromatic (PAN) image is known as the pansharpening, which is a very important preprocessing step in remote sensing. Component substitution (CS)-based methods such as intensity-hue-saturation (IHS) may be the most widely used pansharpening methods and have been adopted in most of the professional software. However, the CS-based methods generally suffer from severe spectral distortions. In this paper, we propose a post-processing technique called guided residual fusion (GRF) to compensate the results of CS-based methods, which inherits high spatial fidelity from generalized IHS (GIHS, take it as an example) and greatly alleviates the spectral distortions. The main idea of our method is to estimate a residual map between the result of GIHS and the ideal high-resolution multispectral (HRMS) image. Specifically, we first obtain an approximate HRMS by inserting the pixels of the MS image according to the spatial resolution ratio between the PAN and MS. Most of the pixels in this HRMS are unknowns. We can get a masked residual map by subtracting the known values of HRMS and their corresponding counterparts of the result of GIHS. Then, we use an interpolation filter and an edge-preserving image smoothing filter to efficiently fill missing pixel values in the masked residual map, obtaining an accurate residual map. Extensive experiments on six real datasets (including datasets acquired by the Geoeye1, IKONOS, Landsat-8, QuickBird, SPOT6, and Worldview-3 sensors) demonstrate that our method outperforms the compared state-of-the-art methods, that is, the performance of our method ranks the best among all nine methods on three datasets (Geoeye1, IKONOS, and SPOT6), the second on two datasets (QuickBird and Worldview-3), and the third on the Landsat-8 dataset. The source code will be made public in https://ljy-rs.github.io/web.

Noise Reduction Methods for Charge Stability Diagrams of Double Quantum Dots

Operating semiconductor quantum dots as quantum bits requires isolating single electrons by adjusting gate voltages. The transitions of electrons to and from the dots appear as a honeycomb-like pattern in recorded charge stability diagrams (CSDs). Thus, detecting the pattern is essential to tune a double dot, but manual tuning is seriously time-consuming. However, automation of this process is difficult because the transitions’ contrast is often low, and noise and background disorder potential shifts disturb the CSDs. Therefore, the signal-to-noise ratio needs to be increased to improve the detection of the line pattern. For this purpose, we evaluate a representative set of edge-preserving smoothing filters and compare them both quantitatively and qualitatively by suitable metrics and visual assessment. We generate artificial data to use full-reference metrics for the evaluation procedure and to optimize the filter parameters. Based on the results of this article, the methods attain a moderate to good amount of noise reduction and structure improvement dependent on the different CSD qualities. In conclusion, we suggest introducing the block-matching and three-dimensional transform-domain filter into the automated tuning processing pipeline. If the data are corrupted by significant amounts of random telegraph noise, the bilateral filter and the rolling guidance filter are also good choices.

A field programmable gate array-based biomedical noise reduction framework using advanced trilateral filter

Filtering is one of the essential tools utilized to remove undesirable features in biomedical images. Most biomedical image denoising systems are used for clinical diagnosis. So, in this paper, we use the advanced trilateral filter in the field programmable gate array (FPGA) for removing noise in the biomedical image. Generally, the trilateral filter is used as an edge preserving smoothing filter. This advanced approach of trilateral filter gives the best noise diminution and enhances the image quality. This paper also proposes the hardware implementation of an efficient FPGA-based advanced trilateral filter on real time execution. In this manuscript, we intend to design and implement the FPGA architecture using an advanced trilateral filter. Biomedical images with different noises are used during implementation and compared with the existing bilateral and trilateral architecture to assess the proposed architecture performance. For evaluating the performance metrics of the proposed advanced trilateral filter on MATLAB platform, peak signal-to-noise ratio (PSNR), mean squared error (MSE) and structured similarity index (SSIM) are calculated for different biomedical images – such as brain (MRI), chest (x-ray) and lungs (CT) – with different noises – such as salt and pepper, Gaussian, Poisson and Speckle noises – compared with existing bilateral filter and trilateral filter, respectively. The proposed advanced trilateral filter implementations are checked on Virtex-6, Virtex-7 and Zynq FPGA development board using Verilog programming language in Xilinx ISE 14.5 design tools. The simulation outcomes display that the FPGA execution of the advanced trilateral filter contains better noise removal efficiency in biomedical images compared with the existing bilateral and trilateral filter.

Enhancement and Sharpening the Migration Images of the Gravity Field and Its Gradients

Potential field migration represents a rapid technique for imaging the subsurface based on gravity data. However, migration transformation usually produces a smooth and unfocused image of the targets due to the diffusive nature of the potential fields. In this paper, we introduce a method of the migration image enhancement and sharpening based on the application of the hybrid focusing stabilizer, which combines the edge-preserving smoothing filter with the minimum support functional. The method is based on the model resolution matrix of the migration operator. We also improve the migration image with a novel target-oriented migration method. The developed method of migration image enhancement and sharpening is illustrated by synthetic model studies and case studies. The case study involves imaging the full tensor gravity gradient data collected in the Nordkapp Basin of the Barents Sea.

Fast Edge Preserving 2D Smoothing Filter Using Indicator Function

Edge-preserving smoothing filter smoothes the textures while it preserves the information of sharp edges. In image processing, this filter is used as a fundamental process of many applications. In this paper, we propose a new approach for edge-preserving smoothing filter. Our method uses 2D filter to smooth images and we apply indicator function to restrict the range of filtered pixels for edge-preserving. To define the indicator function, we recalculate the distance between each pixel by using edge information. The nearby pixels in the new domain are used for smoothing. Since our method constrains the pixels used for filtering, its implementation is quite fast. We demonstrate the usefulness of our new edge-preserving smoothing method for some applications.

Including edge preserving smoothing filter within blocky- constrained, target-oriented AVA inversions

Resolving thin layers and achieve focused layer boundaries is one of the major challenges in seismic inversion. This translates into recovering a blocky solution with sparse spatial derivatives of model parameters. Here, we present two iterative focusing regularization techniques for target-oriented Amplitude Versus Angle (AVA) inversion. Target-oriented means that only the AVA responses of the reflections of interest are inverted for the simultaneous estimation of P-wave, S-wave and density reflectivities. The first approach imposes Cauchy constraints on the spatial model derivatives whereas, the second is inspired by the minimum gradient support regularization. Both the implemented algorithms enhance their focusing and edge-preserving abilities by exploiting an Edge Preserving Smoothing (EPS) filter that is used to compute both the model constraints and the model update. We include a-priori model information into the inversion kernel to guide the convergence of the algorithms toward physically plausible solutions. The two approaches are compared against the standard Bayesian inversion that simply considers Gaussian distributed model parameters, and with the well-known edge-preserving method that assumes Cauchy-distributed derivative of model parameters. For the lack of available field seismic data, we limit the attention to synthetic inversion experiments in which we simulate different signal-to noise (S/N) ratios. The inversion tests prove the suitability of the two proposed approaches for target-oriented AVA inversion and demonstrate their focusing and anti-noise abilities. In particular, the two implemented algorithms outperform the standard Bayesian inversion and the Cauchy approach in cases of low S/N ratios. The two implemented methods are also extremely flexible and can be applied to other linear or non-linear geophysical inverse problems.

Generation of high dynamic range illumination from a single image for the enhancement of undesirably illuminated images

This paper presents an algorithm that enhances undesirably illuminated images by generating and fusing multi-level illuminations from a single image. The input image is first decomposed into illumination and reflectance components by using an edge-preserving smoothing filter. Then the reflectance component is scaled up to improve the image details in bright areas. The illumination component is scaled up and down to generate several illumination images that correspond to certain camera exposure values different from the original. The virtual multi-exposure illuminations are blended into an enhanced illumination, where we also propose a method to generate appropriate weight maps for the tone fusion. Finally, an enhanced image is obtained by multiplying the equalized illumination and enhanced reflectance. Experiments show that the proposed algorithm produces visually pleasing output and also yields comparable objective results to the conventional enhancement methods, while requiring modest computational loads.

Seismic inversion with adaptive edge-preserving smoothing preconditioning on impedance model

Poststack seismic impedance inversion is an effective approach for reservoir prediction. Due to the sensitivity to noise and the oscillation near the bed boundary, Gaussian distribution constrained seismic inversion is unfavorable to delineate the subtle-reservoir and small-scale geologic features. To overcome this shortcoming, we have developed a new method that incorporates a priori knowledge in the seismic inversion through a preconditioning impedance model using the adaptive edge-preserving smoothing (Ad-EPS) filter. The Ad-EPS filter preconditioned impedance model for a blocky solution makes the formation interfaces and geologic edges more precise and sharper in the inverted impedance results and keeps the inversion procedure robust even if random noise exists in the seismic data. Furthermore, compared with the conventional EPS filter, the Ad-EPS filter is able to resolve thick and thin geologic features through window size scanning, which is used to find the best-fitting window size for each sample to be filtered. The results of numerical examples and real seismic data test indicate that our inversion method can suppress noise to obtain a “blocky” inversion result and preserve small geologic features.

An empirical correction method for beam-hardening artifact in Computerized Tomography (CT) images

A new empirical method has been proposed for the suppression of beam hardening artifacts. It utilizes discrete cosine transform based missing data estimation and an edge-preserving smoothing filter. The method does not need any information about the experiment specification or the X-ray spectrum used and the material being inspected. The corrections are done in the image domain. The proposed method is first tested on three test samples. It is then applied to the cylindrical sample made of Perspex, aluminiumand steel. The results are compared with that of the existing work and the outcome is promising. The horizontal line profile and other parametric comparisons indicate that the proposed method can suppress beam-hardening artifacts significantly. The application of Sobolev space error analysis has also been made to ensure good global reconstruction.

Propagated guided image filtering for edge-preserving smoothing

This paper proposes an edge-preserving smoothing filtering algorithm based on guided image filter (GF). GF is a well-known edge-preserving smoothing filter, but is ineffective in certain cases. The proposed GF enhancement provides a better solution for various noise levels associated with image degradation. In addition, halo artifacts, the main drawback of GF, are well suppressed using the proposed method. In our proposal, linear GF coefficients are updated sequentially in the spatial domain by using a new cost function, whose solution is a weighted average of the neighboring coefficients. The weights are determined differently depending on whether the pixels belong to the edge region, and become zero when a neighborhood pixel is located within a region separated from the center pixel. This propagation procedure is executed twice (from upper-left to lower-right, and vice versa) to obtain noise-free edges. Finally, the filtering output is computed using the updated coefficient values. The experimental results indicate that the proposed algorithm preserves edges better than the existing algorithms, while reducing halo artifacts even in highly noisy images. In addition, the algorithm is less sensitive to user parameters compared to GF and other modified GF algorithms.

Edge-preserving smoothing for simultaneous-source full-waveform inversion model updates in high-contrast velocity models

Full-waveform inversion (FWI) can provide accurate estimates of subsurface model parameters. In spite of its success, the application of FWI in areas with high-velocity contrast remains a challenging problem. Quadratic regularization methods are often adopted to stabilize inverse problems. Unfortunately, edges and sharp discontinuities are not adequately preserved by quadratic regularization techniques. Throughout the iterative FWI method, an edge-preserving filter, however, can gently incorporate sharpness into velocity models. For every point in the velocity model, edge-preserving smoothing assigns the average value of the most uniform window neighboring the point. Edge-preserving smoothing generates piecewise-homogeneous images with enhanced contrast at boundaries. We adopt a simultaneous-source frequency-domain FWI, based on quasi-Newton optimization, in conjunction with an edge-preserving smoothing filter to retrieve high-contrast velocity models. The edge-preserving smoothing filter gradually removes the artifacts created by simultaneous-source encoding. We also have developed a simple model update to prevent disrupting the convergence of the optimization algorithm. Finally, we perform tests to examine our algorithm.

AUTOMATED SCALING REGION OF INTEREST WITH ITERATIVE EDGE PRESERVING IN FORWARD-BACKWARD TIME-STEPPING

The thesis reported on a deterministic inverse scattering method in time-domain to reconstruct dielectric profiles of an unknown embedded object within its peripheral region. Image reconstruction of geometrically simple objects taken after breast profiles and lung(s) model are solely done by simulation executed in single computing. In effort to alleviate the nonlinearity problem of inverse scattering, the inversion technique has been integrated with varied types of techniques to improve the inversion solution. The extended algorithm would only increase the computational cost, nevertheless would never eliminate the nonlinearity problem that degrade the inversion solution. The reconstruction process started in a coarse region which then rescaled down according to object geometry configuration. This is accomplished by combining an inversion technique of Forward-Backward Time-Stepping (FBTS) with an Automated Scaling Region of Interest (AS-ROI) method. Edge preserving techniques comprises of edge preserving regularization and anisotropic diffusion are integrated into the combined FBTS and AS-ROI to further increase the accuracy level in the profiles’ intensity. Accuracy of reconstructed object is validated by using mean squared error (MSE), relative error (RE) and Euclidean distance (ED) that measure the precision in terms of pixels’ intensity, size and localization. Results exhibited significant improvement in the accuracy level of reconstructed images with a combined method of FBTS and AS-ROI. AS-ROI has successfully increased the pixels precision in relative to FBTS about 10.33% for breast model and 25.17% for lung model. The accuracy increment by AS-ROI is due to better fields penetration as exterior pixels are replaced with background layer of low profiles. In the combined rescaled and regularized method in FBTS, edge preserving smoothing filter and regularization are alternately imposed on the improved reconstructed profiles by AS-ROI. Therefore, the accuracy is further increased to 18.58% and 40.68%, respectively for breast and lung model. In term of size estimation, average error in object’s radius is analogous to accuracy level measured in MSE. It indicates that efficiency of AS-ROI is highly relied upon the accuracy of FBTS estimation in reconstructing image profiles prior to rescaling process. Apart from that, accuracy in size estimation differ for varied shapes due to number of pixels at the boundary. Average RE is 61.94% for a circular shape in breast model, meanwhile attains RE of 4.17% for a U-shape object. The highest RE for a single circular tumour’s size in lung model is 62.5%. However, object localization by AS-ROI is 100% for a circular and U-shape object in breast model. Nevertheless, AS-ROI attains an average ED of 2.3 for lung model. Computational time decreases accordingly with the reduced number of pixels involved after the rescaling process. Reduction in the computational time is 13.06% for breast model, nonetheless 28.74% for lung model. Significant time reduction observed for lung model is benefited from considerable number of pixels that has been removed from the original image. The inclusion of edge preserving techniques into the combined AS-ROI with FBTS however has slightly increase the computational time about 5.87% and 4.69% for breast and lung model, respectively. Nevertheless, it compensates the computational cost for higher accuracy of image profiles’ intensities.

Single Image De-Hazing Using Globally Guided Image Filtering.

Local edge-preserving smoothing techniques such as guided image filtering (GIF) and weighted guided image filtering (WGIF) could not preserve fine structure. In this paper, a new globally guided image filtering (G-GIF) is introduced to overcome the problem. The G-GIF is composed of a global structure transfer filter and a global edge-preserving smoothing filter. The proposed filter is applied to study single image haze removal. Experimental results show that fine structure of the dehazed image is indeed preserved better by the proposed G-GIF and the dehazed images by the proposed G-GIF are sharper than those dehazed images by the existing GIF.

Relative reductive structure-aware regression filter

Structure-aware image smoothing is a challenging and significant technique to remedy the limitation of current edge-preserving smoothing filters for extracting the prominent structures. To improve the technique, we propose a novel structure-aware filter via bilateral kernel regression with a variational structure-kernel descriptor. First, the relative reductive texture decomposition is applied to construct the structure-kernel descriptor. Then, the descriptor is incorporated into the bilateral kernel regression to achieve an expected structure preservation output. Algorithmically, a close-form numerically iterative solver is exploited to achieve the efficient and effective implementation. At last, some experimental self-evaluations and visual applications are presented to demonstrate that our method leads to better performance than the state-of-the-art solutions.

Feasibility of patient dose reduction based on various noise suppression filters for cone-beam computed tomography in an image-guided patient positioning system

We investigated the feasibility of patient dose reduction based on six noise suppression filters for cone-beam computed tomography (CBCT) in an image-guided patient positioning (IGPP) system. A midpoint dose was employed as a patient dose index. First, a reference dose (RD) and low-dose (LD)-CBCT images were acquired with a reference dose and various low doses. Second, an automated rigid registration was performed for three axis translations to estimate patient setup errors between a planning CT image and the LD-CBCT images processed by six noise suppression filters (averaging filter, median filter, Gaussian filter, edge-preserving smoothing filter, bilateral filter, and adaptive partial median filter (AMF)). Third, residual errors representing the patient positioning accuracy were calculated as Euclidean distances between the setup error vectors estimated using the LD-CBCT and RD-CBCT images. Finally, the residual errors as a function of the patient dose index were estimated for LD-CBCT images processed by six noise suppression filters, and then the patient dose indices for the filtered LD-CBCT images were obtained at the same residual error as the RD-CBCT image. This approach was applied to an anthropomorphic phantom and four cancer patients. The patient dose for the LD-CBCT images was reduced to 19% of that for the RD-CBCT image for the phantom by using AMF, while keeping a same residual error of 0.47 mm as the RD-CBCT image by applying the noise suppression filters to the LD-CBCT images. The average patient dose was reduced to 31.1% for prostate cancer patients, and it was reduced to 82.5% for a lung cancer patient by applying the AMF. These preliminary results suggested that the proposed approach based on noise suppression filters could decrease the patient dose in IGPP systems.

Terracing gravity and magnetic data using edge-preserving smoothing filters

One major purpose of gravity and magnetic transformations is to produce a result that can be related to geology. The terracing operator achieves this purpose by converting gravity and magnetic data into a geologic map-like field wherein homogeneous domains with sharp domain boundaries are defined. Edge-preserving smoothing filters developed in image processing have the same capability. I have applied the Kuwahara, mean of least variance, and symmetric nearest neighbor filters to gravity and magnetic data. Synthetic and field data examples suggest that these edge-preserving smoothing filters produce terraced effects cleaner than the terracing operator, and the mean of least variance filter often produces the cleanest and sharpest result.

Terracing gravity and magnetic data using edge-preserving smoothing filters

Terracing gravity and magnetic data using edge-preserving smoothing filters