Perceptual Sharpness Research Articles

Investigation of a Robust Blind Deconvolution Algorithm Using Extracted Structures in Light Microscopy Images of Salivary Glands: A Pilot Study

Although light microscopy (LM) images are widely used to observe various bodily tissues, including salivary glands, reaching a satisfactory spatial resolution in the final images remains a major challenge. The objective of this study was to model a robust blind deconvolution algorithm using the extracted structure and analyze its applicability to LM images. Given LM images of the salivary glands, the proposed robust blind deconvolution method performs non-blind deconvolution after estimating the structural map and kernel of each image. To demonstrate the usefulness of the proposed algorithm for LM images, the perceptual sharpness index (PSI), Blanchet’s sharpness index (BSI), and natural image quality evaluator (NIQE) were used as evaluation metrics. We demonstrated that when the proposed algorithm was applied to salivary gland LM images, the PSI and BSI were improved by 7.95% and 7.44%, respectively, compared with those of the conventional TV-based algorithm. When the proposed algorithm was applied to an LM image, we confirmed that the NIQE value was similar to that of a low-resolution image. In conclusion, the proposed robust blind deconvolution algorithm is highly applicable to salivary gland LM images, and we expect that further applications will become possible.

A skewness reformed complex diffusion based unsharp masking for the restoration and enhancement of Poisson noise corrupted mammograms

• A skewness reformed complex diffusion based unsharp masking is proposed for removal of quantum noise which follows Poisson distribution, enhancement of abnormalities, edge preservation and restoration of information in a single framework. • In the proposed method a smoothing filter is swapped with a skewness reformed complex diffusion in unsharp masking. • To improve the efficacy of proposed method, the edge threshold constraint of diffusion coefficient has been automated by the help of skewness of image instead of user-based value. It helps to detect the sharp details, boundary and edges of image. • In order to judge the proficiency of the proposed work, MIAS, Benign Breast Tumor IEEE dataset and phantom breast mammogram images are used for the qualitative and quantitative analysis. • The proposed method is compared with other state of the art with respect to qualitative as well as quantitative analysis. • The quantitative analysis is measured in terms of mean squared error, peak signal to noise ratio, correlation parameter, normalized absolute error, universal quality index, structural similarity index, measure of enhancement, measure of enhancement by entropy, second derivative like measurement, average gradient and perceptual sharpness index respectively. Mammography is a proven imaging modality for the screening of breast cancer that helps to evaluate the existence of calcification, masses, tissue density, lump shape and edges. These features are generally corrupted by noise, distortion and artifacts during the image acquisition. The presence of these variability reduces the quality of image and limits the efficacy of radiologist’s interpretation. To overcome these barriers, a skewness reformed complex diffusion based unsharp masking is proposed, where smoothing filter is changed with a skewness reformed complex diffusion (SRCD) in unsharp masking. To improve the efficacy of SRCD, edge threshold constraint of the diffusion coefficient has been automated instead of manual calculation. It has been automated by the skewness calculation of the image. It helps to detect the fine details, boundary and edges of image. In order to judge the proficiency of the proposed method, Mammography Image Analysis Society, Benign Breast Tumor IEEE dataset and phantom mammogram images are casted-off. The proposed method is compared with other state of the art with respect to qualitative and quantitative analysis. The quantitative analysis is measured in terms of mean squared error, peak signal to noise ratio, correlation parameter, normalized absolute error, universal quality index, structural similarity index, measure of enhancement, measure of enhancement by entropy, second derivative like measurement, average gradient and perceptual sharpness index. The obtained result confines the effectiveness of the proposed method and it provides removal of Poisson noise, enhancement of abnormalities, edge preservation and restoration of information in a single framework.

Deep feature loss to denoise OCT images using deep neural networks.

.Significance: Speckle noise is an inherent limitation of optical coherence tomography (OCT) images that makes clinical interpretation challenging. The recent emergence of deep learning could offer a reliable method to reduce noise in OCT images.Aim: We sought to investigate the use of deep features (VGG) to limit the effect of blurriness and increase perceptual sharpness and to evaluate its impact on the performance of OCT image denoising (DnCNN).Approach: Fifty-one macula-centered OCT pairs were used in training of the network. Another set of 20 OCT pair was used for testing. The DnCNN model was cascaded with a VGG network that acted as a perceptual loss function instead of the traditional losses of and . The VGG network remains fixed during the training process. We focused on the individual layers of the VGG-16 network to decipher the contribution of each distinctive layer as a loss function to produce denoised OCT images that were perceptually sharp and that preserved the faint features (retinal layer boundaries) essential for interpretation. The peak signal-to-noise ratio (PSNR), edge-preserving index, and no-reference image sharpness/blurriness [perceptual sharpness index (PSI), just noticeable blur (JNB), and spectral and spatial sharpness measure (S3)] metrics were used to compare deep feature losses with the traditional losses.Results: The deep feature loss produced images with high perceptual sharpness measures at the cost of less smoothness (PSNR) in OCT images. The deep feature loss outperformed the traditional losses ( and ) for all of the evaluation metrics except for PSNR. The PSI, S3, and JNB estimates of deep feature loss performance were 0.31, 0.30, and 16.53, respectively. For L1 and L2 losses performance, the PSI, , and JNB were 0.21 and 0.21, 0.17 and 0.16, and 14.46 and 14.34, respectively.Conclusions: We demonstrate the potential of deep feature loss in denoising OCT images. Our preliminary findings suggest research directions for further investigation.

Reducing Contrast Agent Dose in Cardiovascular MR Angiography with Deep Learning.

Contrast-enhanced magnetic resonance angiography (MRA) is used to assess various cardiovascular conditions. However, gadolinium-based contrast agents (GBCAs) carry a risk of dose-related adverse effects. To develop a deep learning method to reduce GBCA dose by 80%. Retrospective and prospective. A total of 1157 retrospective and 40 prospective congenital heart disease patients for training/validation and testing, respectively. A 1.5 T, T1-weighted three-dimensional (3D) gradient echo. A neural network was trained to enhance low-dose (LD) 3D MRA using retrospective synthetic data and tested with prospective LD data. Image quality for LD (LD-MRA), enhanced LD (ELD-MRA), and high-dose (HD-MRA) was assessed in terms of signal-to-noise ratio (SNR), contrast-to-noise ratio (CNR), and a quantitative measure of edge sharpness and scored for perceptual sharpness and contrast on a 1-5 scale. Diagnostic confidence was assessed on a 1-3 scale. LD- and ELD-MRA were assessed against HD-MRA for sensitivity/specificity and agreement of vessel diameter measurements (aorta and pulmonary arteries). SNR, CNR, edge sharpness, and vessel diameters were compared between LD-, ELD-, and HD-MRA using one-way repeated measures analysis of variance with post-hoc t-tests. Perceptual quality and diagnostic confidence were compared using Friedman's test with post-hoc Wilcoxon signed-rank tests. Sensitivity/specificity was compared using McNemar's test. Agreement of vessel diameters was assessed using Bland-Altman analysis. SNR, CNR, edge sharpness, perceptual sharpness, and perceptual contrast were lower (P< 0.05) for LD-MRA compared to ELD-MRA and HD-MRA. SNR, CNR, edge sharpness, and perceptual contrast were comparable between ELD and HD-MRA, but perceptual sharpness was significantly lower. Sensitivity/specificity was 0.824/0.921 for LD-MRA and 0.882/0.960 for ELD-MRA. Diagnostic confidence was 2.72, 2.85, and 2.92 for LD, ELD, and HD-MRA, respectively (PLD-ELD , PLD-HD < 0.05). Vessel diameter measurements were comparable, with biases of 0.238 (LD-MRA) and 0.278 mm (ELD-MRA). Deep learning can improve contrast in LD cardiovascular MRA. 2 TECHNICAL EFFICACY: Stage 2.

Light Field All-in-focus Image Fusion Based on Edge Enhanced Guided Filtering

Affected by the micro-lens geometric calibration accuracy of the light field camera, the decoding error of the 4D light field in the angular direction will cause the edge information loss of the integrated refocused image, which will reduce the accuracy of the all-in-focus image fusion. In this paper, a light field all-in-focus image fusion algorithm based on edge-enhanced guided filtering is proposed. Through multi-scale decomposition of the digital refocused images and guided filtering optimization of the feature layer decision map, the final all-in-focus image is obtained. Compared with the traditional fusion algorithm, the edge information loss caused by the 4D light field calibration error is compensated in the presented method. In the step of multi-scale decomposition of the refocused image, the edge layer extraction is added to accomplish the high-frequency information enhancement. Then the multi-scale evaluation model is established to optimize the edge layer’s guided filtering parameters to obtain a better light field all-in-focus image. The experimental results show that the edge intensity and the perceptual sharpness of the all-in-focus image can be improved without significantly reducing the similarity between the all-in-focus image and the original image.

Light field all-in-focus image fusion based on spatially-guided angular information

Compared to traditional 2D images, light field images record both spatial and angular information of the scene, which can provide more data for image fusion. In this paper, a light field all-in-focus image fusion algorithm based on spatially-guided angular information is proposed. In the proposed method, the initial weight maps carrying the angular information are calculated by comparing the block variance of the 4D light field data. The initial weight maps are then guided by digital refocused images carrying the spatial information to obtain the refined weight maps. In the refocused image multi-scale decomposition, the micro-lens calibration error is considered and the additional edge layers are extracted to suppress the edge artifacts. Experiments demonstrate the effectiveness of the proposed algorithm. Quantitative evaluation results show that the proposed algorithm performs the best in the feature-based index and structural similarity-based index without sacrificing the information and perceptual sharpness of the fused image. • A light field all-in-focus image fusion algorithm based on guided filtering is proposed. • Using both the spatial and angular information of the light field and combining their advantages in the guided filtering. • Calculating the initial weight maps with higher confidence based on the angular information of the light field. • Considering the micro-lens calibration error and decomposing the additional edge layers to suppress the edge artifact.

Nonlinear sharpening of MR images using a locally adaptive sharpness gain and a noise reduction parameter

Well-defined boundaries are necessary to allow precise delineation of morphological structures from magnetic resonance images. Existing state-of-the-art sharpening techniques like unsharp masking (UM) produce discontinuity artefacts and have multiple operational parameters. Tuning multiple parameters together is cumbersome. A computationally efficient and noise robust nonlinear sharpening scheme which is free from discontinuity and saturation artefacts, with very less number of arbitrary parameters, is proposed in this paper. As an inverse mathematical problem, the sharpened image is computed from the amplified first derivative. To constrain the noise amplification, the local value of the amplification factor is considered as a nonlinear function of local average of absolute directional gradients. The proposed sharpening scheme is compared with two of its best possible alternatives, contrast limited adaptive histogram equalization (CLAHE) and UM in terms of sharpness of the output image, thinness of salient edges, feature preservation, saturation and edge quality degradation due to noise, using perceptual sharpness index (PSI), second-order derivative-based measure of enhancement (SDME), edge model-based blur metric (EMBM), structural similarity index metric (SSIM), peak signal-to-noise ratio (PSNR), saturation evaluation index (SEI) and sharpness of ridges (SOR). The proposed nonlinear sharpening scheme exhibited higher PSI, SDME, PSNR and SSIM and lower EMBM, SOR and computational time, compared to CLAHE and UM. The proposed scheme is found to be superior to the existing state-of-the-art techniques like CLAHE and UM, in terms of sharpness as well as thinness of salient edges in the sharpened image, robustness to noise, discontinuity artefacts, saturation and computational time. Because of minimum number of operational parameters, it is user friendly too.

A Fuzzy Sharpness Metric for Magnetic Resonance Images

Abstract A fuzzy-based sharpness metric for the objective measurement of sharpness of Magnetic Resonance (MR) images is proposed in this paper. In the proposed metric, Quadratic Index of fuzziness (QIF) is used to quantitatively express image sharpness. The proposed metric is found to be superior to Maximum Local Variation (MLV) metric, Perceptual Sharpness Index (PSI), Second order Derivative based Measure of Enhancement (SDME), Blanchet’s Sharpness Index (BSI) and Roffet’s Blur Metric (RBM) in terms of correlation with subjective quality ratings and computational time.

A prospective case study of high boost, high frequency emphasis and two-way diffusion filters on MR images of glioblastoma multiforme.

Glioblastoma multiforme (GBM) appears undifferentiated and non-enhancing on magnetic resonance (MR) imagery. As MRI does not offer adequate image quality to allow visual discrimination of the boundary between GBM focus and perifocal vasogenic edema, surgical and radiotherapy planning become difficult. The presence of noise in MR images influences the computation of radiation dosage and precludes the edge based segmentation schemes in automated software for radiation treatment planning. The performance of techniques meant for simultaneous denoising and sharpening, like high boost filters, high frequency emphasize filters and two-way anisotropic diffusion is sensitive to the selection of their operational parameters. Improper selection may cause overshoot and saturation artefacts or noisy grey level transitions can be left unsuppressed. This paper is a prospective case study of the performance of high boost filters, high frequency emphasize filters and two-way anisotropic diffusion on MR images of GBM, for their ability to suppress noise from homogeneous regions and to selectively sharpen the true morphological edges. An objective method for determining the optimum value of the operational parameters of these techniques is also demonstrated. Saturation Evaluation Index (SEI), Perceptual Sharpness Index (PSI), Edge Model based Blur Metric (EMBM), Sharpness of Ridges (SOR), Structural Similarity Index Metric (SSIM), Peak Signal to Noise Ratio (PSNR) and Noise Suppression Ratio (NSR) are the objective functions used. They account for overshoot and saturation artefacts, sharpness of the image, width of salient edges (haloes), susceptibility of edge quality to noise, feature preservation and degree of noise suppression. Two-way diffusion is found to be superior to others in all these respects. The SEI, PSI, EMBM, SOR, SSIM, PSNR and NSR exhibited by two-way diffusion are 0.0016 ± 0.0012, 0.2049 ± 0.0187, 0.0905 ± 0.0408, 2.64 × 1012 ± 1.6 × 1012, 0.9955 ± 0.0024, 38.214 ± 5.2145 and 0.3547 ± 0.0069, respectively.

Sharpening enhancement technique for MR images to enhance the segmentation

Abstract This study presents a sharpening image enhancement technique based on the Laplacian pyramid (LP) and singular value decomposition (SVD) to improve the visibility and segmentation of subtle organs. The technique utilizes the shape-invariant properties of LP and the SVD techniques to enhance the perceptual sharpness of an image. The sharpening enhancement of magnetic resonance images not only sharpens the edges, but also reduces the noise effect. The results are compared with state-of-art enhancement techniques. The performance measures like peak-signal-to noise (PSNR), mean structural similarity index (MSSIM), and absolute mean brightness error (AMBE) evaluates the improved performance of the proposed technique.

GPU-based real-time super-resolution system for high-quality UHD video up-conversion

Super-resolution (SR) is a technique that reconstructs high-resolution images using the information present in low-resolution images. Due to their potentials of being used in wide range of image and video applications, various SR algorithms have been studied and proposed in the literature until recently. However, many of the algorithms provide insufficient perceptual quality, possess high computational complexity, or have high memory requirement, which make them hard to apply on consumer-level products. Therefore, in this paper we propose an effective super-resolution method that not only provides an excellent visual quality but also a high-speed performance suitable for video conversion applications. The proposed super-resolution adopts self-similarity framework, which reconstructs the high-frequency (HF) information of the high-resolution image by referring to the image pairs generated from self-similar regions. The method further enhances the perceptual sharpness of the video through region-adaptive HF enhancement algorithm and applies iterative back projection to maintain its consistency with the input image. The proposed method is suitable for parallel processing and therefore is able to provide its superb visual quality on a high conversion speed through GPU-based acceleration. The experimental results show that the proposed method has superior HF reconstruction performance compared to other state-of-the-art upscaling solutions and is able to generate videos that are visually as sharp as the original high-resolution videos. On a single PC with four GPUs, the proposed method can convert Full HD resolution video into UHD resolution with real-time conversion speed. Due to its fast and high-quality conversion capability, the proposed method can be applied on various consumer products such as UHDTV, surveillance system, and mobile devices.

Blind Sharpness Prediction for Ultrahigh-Definition Video Based on Human Visual Resolution

We explore a no-reference sharpness assessment model for predicting the perceptual sharpness of ultrahigh-definition (UHD) videos through analysis of visual resolution variation in terms of viewing geometry and scene characteristics. The quality and sharpness of UHD videos are influenced by viewer perception of the spatial resolution afforded by the UHD display, which depends on viewing geometry parameters including display resolution, display size, and viewing distance. In addition, viewers may perceive different degrees of quality and sharpness according to the statistical behavior of the visual signals, such as the motion, texture, and edge, which vary over both spatial and temporal domains. The model also accounts for the resolution variation associated with fixation and foveal regions, which is another important factor affecting the sharpness prediction of UHD video over the spatial domain and which is caused by the nonuniform distribution of the photoreceptors. We calculate the transition of the visually salient statistical characteristics resulting from changing the display’s screen size and resolution. Moreover, we calculated the temporal variation in sharpness over consecutive frames in order to evaluate the temporal sharpness perception of UHD video. We verify that the proposed model outperforms other sharpness models in both spatial and temporal sharpness assessments.

Perceptual sharpness metric based on human visual system

Sharpness is an important quality attribute influencing image quality. Existing sharpness assessment models have not given sufficient consideration to the human visual system, especially the human luminance masking effect. On the basis of an isotropic local contrast model, an objective perceptual sharpness assessment model is proposed combined with the human luminance masking effect. Experimental results show that the proposed perceptual sharpness metric provides better predictions than four existing sharpness metrics and can evaluate image perceptual sharpness effectively.

A Perceptual Image Sharpness Metric Based on Local Edge Gradient Analysis

In this letter, a no-reference perceptual sharpness metric based on a statistical analysis of local edge gradients is presented. The method takes properties of the human visual system into account. Based on perceptual properties, a relationship between the extracted statistical features and the metric score is established to form a Perceptual Sharpness Index (PSI). A comparison with state-of-the-art metrics shows that the proposed method correlates highly with human perception and exhibits low computational complexity. In contrast to existing metrics, the PSI performs well for a wide range of blurriness and shows a high degree of invariance for different image contents.

No-reference sharpness metric based on inherent sharpness

Proposed is a no-reference sharpness metric that employs the concept of inherent sharpness. It provides a perceptual sharpness score based on wavelet coefficients. The proposed metric is highly correlated with subjective sharpness evaluations and outperforms other state-of-the-art metrics.

Sharpness preserving image enlargement by using self-decomposed codebook and Mahalanobis distance

We propose an image enlargement method preserving perceptual sharpness, which is achieved by augmenting a low resolution image with high-frequency components from a given image itself. The estimation of high-frequency image components is performed by a codebook built by a decomposition of the given image, i.e. a self-decomposed codebook. The rational that is exploited in this approach is the shape-invariant properties of edges across scales. As a distance measure for matching from the codebook, we employ the Mahalanobis distance which is a local distance measure incorporating pixel correlation. The effectiveness of the proposed method is verified by some image enlargement experiments. Consequently, the experimental results show that the performance of the proposed method is superior to other conventional image enlargement methods objectively and subjectively.

Image enhancement by nonlinear extrapolation in frequency space

A technique for enhancing the perceptual sharpness of an image is described. The enhancement algorithm augments the frequency content of the image using shape-invariant properties of edges across scale by using a nonlinearity that generates phase coherent higher harmonics. The procedure utilizes the Laplacian transform and the Laplacian pyramid image representation. Results are presented depicting the power-spectra augmentation and the visual enhancement of several images. Simplicity of computations and ease of implementation allow for real-time applications such as high-definition television (HDTV).