Objective Image Quality Measures Research Articles

Visual-PSNR measure of image quality

Objective assessment of image quality is important in numerous image and video processing applications. Many objective measures of image quality have been developed for this purpose, of which peak signal-to-noise ratio PSNR is one of the simplest and commonly used. However, it sometimes does not match well with objective mean opinion scores (MOS). This paper presents a novel objective full-reference measure of image quality (VPSNR), which is a modified PSNR measure. It will be shown that VPSNR takes into account some features of the human visual system (HVS). The performance of VPSNR is validated using a data set of four image databases, and in this article it is shown that for images compressed by block-based compression algorithms (like JPEG) the proposed measure in the pixel domain matches well with MOS.

The effect of iterative reconstruction on quantitative computed tomography assessment of coronary plaque composition

To compare coronary plaque size and composition as well as degree of coronary artery stenosis on coronary Computed Tomography angiography (CCTA) using three levels of iterative reconstruction (IR) with standard filtered back projection (FBP). In 63 consecutive patients with a clinical indication for CCTA 55 coronary plaques were analysed. Raw data were reconstructed using standard FBP and levels 2, 4 and 6 of a commercially available IR algorithm (iDose(4)). CT attenuation and noise were measured in the aorta and two coronary arteries. Both signal-to-noise-ratio (SNR) and contrast-to-noise ratio (CNR) were calculated. The amount of lipid, fibrous and calcified plaque components and mean cross-sectional luminal area were analysed using dedicated software. Image noise was reduced by 41.6% (p < 0.0001) and SNR and CNR in the aorta were improved by 73.4% (p < 0.0001) and 72.9% (p < 0.0001) at IR level 6, respectively. IR improved objective image quality measures more in the aorta than in the coronary arteries. Furthermore, IR had no significant effect on measurements of plaque volume and cross-sectional luminal area. The application of IR significantly improves objective image quality, and does not alter quantitative analysis of coronary plaque volume, composition and luminal area.

영상 크기 변환에서 화소들의 생략 및 추정 위치를 고려한 객관적 영상 화질 측정

Image scaling is used for a variety of real-life applications. In order to evaluate the performance of transform functions, the image quality are compared together before and after processing. For the objective evaluation of the transform functions, the exact criterion of image quality is required, and various aspects approaches are practically performed. However, few researches have been conducted on image quality measurement considering the position of pixels that are skipped or newly generated in the process of the image scaling. Therefore this paper focuses on the objective image quality measurement for positions of skipped or estimated pixels in the image scaling. The proposed method generated new image quality measure considering the positional changes using a conventional measure and evaluated sensitivity about positional changes. Through this experiments, it is observed that conventional image quality measurement is definitely affected by positional changes of a skipped and estimated pixels. It is also confirmed that the proposed method is an objective criterion to represent image quality for positional changes of skipped or estimated pixels. The proposed method can be used as a criterion to evaluate the performance of image restoration or enhancement functions.

Abdominal CT with model-based iterative reconstruction (MBIR): initial results of a prospective trial comparing ultralow-dose with standard-dose imaging.

The purpose of this study was to report preliminary results of an ongoing prospective trial of ultralow-dose abdominal MDCT. Imaging with standard-dose contrast-enhanced (n = 21) and unenhanced (n = 24) clinical abdominal MDCT protocols was immediately followed by ultralow-dose imaging of a matched series of 45 consecutively registered adults (mean age, 57.9 years; mean body mass index, 28.5). The ultralow-dose images were reconstructed with filtered back projection (FBP), adaptive statistical iterative reconstruction (ASIR), and model-based iterative reconstruction (MBIR). Standard-dose series were reconstructed with FBP (reference standard). Image noise was measured at multiple predefined sites. Two blinded abdominal radiologists interpreted randomly presented ultralow-dose images for multilevel subjective image quality (5-point scale) and depiction of organ-based focal lesions. Mean dose reduction relative to the standard series was 74% (median, 78%; range, 57-88%; mean effective dose, 1.90 mSv). Mean multiorgan image noise for low-dose MBIR was 14.7 ± 2.6 HU, significantly lower than standard-dose FBP (28.9 ± 9.9 HU), low-dose FBP (59.2 ± 23.3 HU), and ASIR (45.6 ± 14.1 HU) (p < 0.001). The mean subjective image quality score for low-dose MBIR (3.0 ± 0.5) was significantly higher than for low-dose FBP (1.6 ± 0.7) and ASIR (1.8 ± 0.7) (p < 0.001). Readers identified 213 focal noncalcific lesions with standard-dose FBP. Pooled lesion detection was higher for low-dose MBIR (79.3% [169/213]) compared with low-dose FBP (66.2% [141/213]) and ASIR (62.0% [132/213]) (p < 0.05). MBIR shows great potential for substantially reducing radiation doses at routine abdominal CT. Both FBP and ASIR are limited in this regard owing to reduced image quality and diagnostic capability. Further investigation is needed to determine the optimal dose level for MBIR that maintains adequate diagnostic performance. In general, objective and subjective image quality measurements do not necessarily correlate with diagnostic performance at ultralow-dose CT.

The optimal dose reduction level using iterative reconstruction with prospective ECG-triggered coronary CTA using 256-slice MDCT

AimTo assess the image quality (IQ) of an iterative reconstruction (IR) technique (iDose4) from prospective electrocardiography (ECG)-triggered coronary computed tomography angiography (coronary CTA) on a 256-slice multi-detector CT (MDCT) scanner and determine the optimal dose reduction using IR that can provide IQ comparable to filtered back projection (FBP). Method and materials110 consecutive patients (69 men, 41 women; age: 54±10 years) underwent coronary CTA on a 256-slice MDCT (Brilliance iCT, Philips Healthcare). The control group (Group A, n=21) were scanned using the conventional tube output (120kVp, 210mAs) and reconstructed using FBP. The other 4 groups were scanned with the same kVp but successively reduced tube output as follows: B[n=15]: 125mAs; C[n=22]: 105mAs; D[n=36]: 84mAs: E[n=16]: 65mAs) and reconstructed using IR levels of L3 (Group B), L4 (Group C) and L5 (Groups D and E), to compensate for the noise increase. All images were reconstructed using the same kernel (XCB). Two radiologists graded IQ in a blinded fashion on a 4-point scale (4 – excellent, 3 – good, 2 – fair and 1 – poor). Quantitative measurements of CT values, image noise and contrast-to-noise (CNR) were measured in each group. A receiver-operating characteristic (ROC) analysis was performed to determine a radiation reduction threshold up to which excellent IQ was maintained. ResultsThere were no significant differences in objective noise, SNR and CNR values among Groups A, B, C, D, and E (P=0.14, 0.09, 0.17, respectively). There were no significant differences in the scores of the subjective IQ between Group A, and Groups B, C, D, E (P=0.23–0.97). Significant differences in image sharpness and study acceptability were observed between groups A and E (P<0.05). Using the criterion of excellent IQ (score 4), the ROC curve of dose levels and IQ acceptability established a reduction of 60% of tube output (Group D) as optimum cutoff point (AUC: 0.72, 95% CI: 0.59–0.86). Group D (84mAs with L5) provided equivalent subjective image ranking (with lumen sharpness taken into account) and objective IQ measurements (noise: 36.5±10.7; SNR: 13.6±4.9; CNR: 16.28±5.4) compared with FBP images in Group A (noise: 35.5±9.4; SNR: 12.4±2.5; CNR: 15.4±3.2) (P=0.14, 0.09, 0.17, respectively). The effective dose (ED) of Group D was 63% lower than that of Group A (1.2±0.1mSv versus 3.2±0.6mSv). ConclusionIterative reconstruction techniques can provide 63% ED reduction in prospectively-triggered coronary CTA using 256-slice MDCT while maintaining excellent image quality.

Effects of Iterative Reconstruction Technique on Image Quality in Cardiac CT Angiography: Initial Experience

Objective: Iterative reconstruction algorithms offer potential radiation dose reduction while maintaining image quality. We attempted to compare image quality parameters of iterative reconstruction and conventional filtered back-projection (FBP) quantitatively in cardiac CT angiography, using second generation dual-source CT technology. Methods: Ten patients were scanned using retrospective ECG-gated dual-source CT for assessment of native coronary artery disease. Multiphase datasets were reconstructed using FBP and iterative reconstruction techniques. Image noise was measured throughout 20 phases of the cardiac cycle (at varying tube currents). Contrast-to-noise ratios (CNR) were compared at 9 specific coronary artery locations in diastole. Results: In all evaluated coronary artery locations, the CNR was significantly improved with iterative reconstruction when compared to conventional FBP (improvement: 39.5 ± 2.8%, p &lt;0.05). Iterative reconstruction demonstrated less image noise across all cardiac phases (reduction: 22.1 ± 4.0%, p &lt;0.05). Conclusions: Iterative reconstruction offers the potential to increase CNR in cardiac CTA. Our experience suggests that iterative reconstruction algorithms have the potential to reduce radiation doses while maintaining similar objective image quality measures such as CNR and noise levels versus standard FBP reconstructions. Further work with comparison to invasive coronary angiography is needed to ensure maintained diagnostic accuracy.

Structural Similarity Measure for Color Images

images reveal more meaningful information to the human observers rather than grayscale ones. Regardless of the advantages of the existing well-known objective image quality measures, one of the common and major limitations of these measures is that they evaluate the quality of grayscale images only and don't make use of color information. In this paper we propose an improved method for image quality assessment that adds a color comparison to the criteria of the well-known Multiscale Structural Similarity index (MSSIM). We evaluated the new color image quality measure through human subjective experiments. Our human subjective evaluation data contains 25 reference images and 875 test images produced by five popular color quantization algorithms. Each of the quantized images was evaluated by twenty two subjects and more than 19200 individual human quality judgments were carried out to obtain the final mean opinion scores. We also tested the proposed method on TID2008 image database to further verify our results. These results indicate that adding color comparison improves MSSIM for many distortions in TID2008 and for assessing quantized images in our database. Keywordsquality assessment, Structural similarity index, Color quantization.

Optimization of Contrast Material Delivery for Dual-Energy Computed Tomography Pulmonary Angiography in Patients With Suspected Pulmonary Embolism

To prospectively compare subjective and objective measures of image quality using 4 different contrast material injection protocols in dual-energy computed tomography pulmonary angiography (CTPA) studies of patients with suspected pulmonary embolism. A total of 100 consecutive patients referred for CTPA for the exclusion of pulmonary embolism were randomized into 1 of 4 contrast material injection protocols manipulating iodine concentration and iodine delivery rate (IDR, expressed as grams of iodine per second): Iomeprol 400 at 3 mL/s (IDR = 1.2 gI/s), iomeprol 400 at 4 mL/s (IDR = 1.6 gI/s), iomeprol 300 at 5.4 mL/s (IDR = 1.6 gI/s), or iomeprol 300 at 4 mL/s (IDR = 1.2 gI/s). Total iodine delivery was held constant. Dual-energy CTPA of the lungs were acquired and used to calculate virtual 120 kV CTPA images as well as iodine perfusion maps. Attenuation values in the thoracic vasculature and image quality of virtual 120 kV CTPAs were compared between groups. Iodine perfusion maps were also compared by identifying differences in the extent of beam-hardening artifacts and subjective image quality. Protocols with an IDR of 1.6 gI/s provided the best attenuation profiles. CTPA image quality was greatest in the high concentration, high IDR (1.6 gI/s) protocol (P < 0.05 for all group comparisons) with no differences between the other groups (all P ≥ 0.05). Extent of beam-hardening artifacts and perfusion map image quality was significantly better using the high concentration, high IDR protocol as compared with all groups (P < 0.05 for all comparisons) and significantly worse using the low concentration, low IDR protocol as compared with all groups (all P ≥ 0.05); no difference was found between the high concentration, low IDR protocol and the low concentration, high IDR protocol (P = 0.73 for comparison of beam-hardening artifacts; P = 0.50 for comparison of perfusion map image quality). High iodine concentration and high IDR contrast material delivery protocols provide the best image quality of both CTPA and perfusion map images of the lung through high attenuation in the pulmonary arteries and minimization of beam-hardening artifacts.

Referenceless image quality evaluation for whole slide imaging

Objective: The image quality in whole slide imaging (WSI) is one of the most important issues for the practical use of WSI scanners. In this paper, we proposed an image quality evaluation method for scanned slide images in which no reference image is required. Methods: While most of the conventional methods for no-reference evaluation only deal with one image degradation at a time, the proposed method is capable of assessing both blur and noise by using an evaluation index which is calculated using the sharpness and noise information of the images in a given training data set by linear regression analysis. The linear regression coefficients can be determined in two ways depending on the purpose of the evaluation. For objective quality evaluation, the coefficients are determined using a reference image with mean square error as the objective value in the analysis. On the other hand, for subjective quality evaluation, the subjective scores given by human observers are used as the objective values in the analysis. The predictive linear regression models for the objective and subjective image quality evaluations, which were constructed using training images, were then used on test data wherein the calculated objective values are construed as the evaluation indices. Results: The results of our experiments confirmed the effectiveness of the proposed image quality evaluation method in both objective and subjective image quality measurements. Finally, we demonstrated the application of the proposed evaluation method to the WSI image quality assessment and automatic rescanning in the WSI scanner.

Image quality assessment in medical imaging based on compressed sensing

Compressed sensing (CS)-based compression and reconstruction of images is an upcoming area of research. In this paper, an efficient application of CS in medical imaging is proposed. The increasing volume of data generated by some medical imaging modalities necessitates different compression techniques for the reduction of storage space and transmission bandwidth. Digital imaging and communications in medicine (DICOM) standard is the backbone of modern medical image display and addresses the exchange of digital information between medical imaging equipment and other systems. When the compression and reconstruction of DICOM images are done using CS, it is found that the artefacts in the reconstructed images are strongly dependent on the sparsifying transforms used in CS. For evaluating the performance of various sparsifying transforms of CS, an objective image quality measurement analysis is conducted. From the experiments, it is found that multi-scale transforms outperform the conventional transforms.

Optimization of hybrid imaging systems based on maximization of kurtosis of the restored point spread function

I propose a novel, but yet simple, no-reference, objective image quality measure based on the kurtosis of the restored point spread function. Using this measure, I optimize several phase masks for extended-depth-of-field in hybrid imaging systems and obtain results that are identical to optimization results based on full-reference image measures of restored images. In comparison with full-reference measures, the kurtosis measure is fast to compute and requires no images, noise distributions, or alignment of restored images, but only the signal-to-noise-ratio.

Subjective image quality assessment based on objective image quality measurement factors

This study was based on the prevailing need to reproduce customers' preferred images. Therefore, we propose a new subjective image quality assessment consisting of languages that customers and manufacturers can easily understand based on the objective image quality measurement factors. In order to examine subjective image quality assessment, these objective image quality measurement factors are expressed by general adjective language throughout the experimental process. To draw adjective languages, the objective image quality measurement factors are presented by stimuli, and subjects expressed their perception of an image quality. The resulting languages are used to question an image quality. According to the analysis, there are meaningful correlations between image quality measurement factors. Moreover, the results indicated that items which strongly affect preference are color reproduction, dynamic range, resolution, and noise, respectively. The study concurs with Engeldrum's The Complete Image Circle and affirms the positive role of interaction between manufacturers and consumers to assess and enhance image quality preference and equipment performance.

Reduced exposure using asymmetric cone beam processing for wide area detector cardiac CT

The purpose of this study was to estimate dose reduction after implementation of asymmetrical cone beam processing using exposure differences measured in a water phantom and a small cohort of clinical coronary CTA patients. Two separate 320×0.5mm detector row scans of a water phantom used identical cardiac acquisition parameters before and after software modifications from symmetric to asymmetric cone beam acquisition and processing. Exposure was measured at the phantom surface with Optically Stimulated Luminescence (OSL) dosimeters at 12 equally spaced angular locations. Mean HU and standard deviation (SD) for both approaches were compared using ROI measurements obtained at the center plus four peripheral locations in the water phantom. To assess image quality, mean HU and standard deviation (SD) for both approaches were compared using ROI measurements obtained at five points within the water phantom. Retrospective evaluation of 64 patients (37 symmetric; 27 asymmetric acquisition) included clinical data, scanning parameters, quantitative plus qualitative image assessment, and estimated radiation dose. In the water phantom, the asymmetric cone beam processing reduces exposure by approximately 20% with no change in image quality. The clinical coronary CTA patient groups had comparable demographics. The estimated dose reduction after implementation of the asymmetric approach was roughly 24% with no significant difference between the symmetric and asymmetric approach with respect to objective measures of image quality or subjective assessment using a four point scale. When compared to a symmetric approach, the decreased exposure, subsequent lower patient radiation dose, and similar image quality from asymmetric cone beam processing supports its routine clinical use.

Example Based Single-Frame Image Super-Resolution by Support Vector Regression

As many other inverse problems, single-frame image super-resolution is an ill-posed problem. The problem has been approached in the context of machine learning. However, the proposed method in this paper is different from other learning based methods regarding how the input/output are formulated as well as how the learning is done. The assumption behind example based methods is the local similarity across seemingly different images. The assumption is illustrated by examples of image coding. Because of the differences in formulating the input/output and the implementation of Support Vector Regression (SVR), it is shown that the proposed approach outperforms the competing SVR method and the kernel regression method in terms of Peak Signal-to-Noise Ratio (PSNR), objective measurements of image quality. Since example based approaches are based on training, in which we know exactly what the output shall be. Therefore, it is proper to objectively measure the performance since the trained model is expected to “correctly” restore the image rather than to enhance the image, e.g. sharpening.

Selecting Low-level Features for Image Quality Assessment by Statistical Methods

Image quality assessment is an important component in every image processing system where the last link of the chain is the human observer. This domain is of increasing interest, in particular in the context of image compression where coding scheme optimization is based on the distortion measure. Many objective image quality measures have been proposed in the literature and validated by comparing them to the Mean Opinion Score (MOS). We propose in this paper an empirical study of several indicators and show how one can improve the performances by combining them. We learn a regularized regression model and apply variable selection techniques to automatically find the most relevant indicators. Our technique enhances the state of the art results on two publicly available databases.

Content-weighted video quality assessment using a three-component image model

Objective image and video quality measures play important roles in numerous image and video processing applications. In this work, we propose a new content-weighted method for full-reference (FR) video quality assessment using a three-component image model. Using the idea that different image regions have different perceptual significance relative to quality, we deploy a model that classifies image local regions according to their image gradient properties, then apply variable weights to structural similarity image index (SSIM) [and peak signal-to-noise ratio (PSNR)] scores according to region. A frame-based video quality assessment algorithm is thereby derived. Experimental results on the Video Quality Experts Group (VQEG) FR-TV Phase 1 test dataset show that the proposed algorithm outperforms existing video quality assessment methods.

New joint demosaicing and zooming algorithm for color filter array

This paper presents a new joint demosaicing and zooming algorithm for digital cameras, each equipped with a single CCD/CMOS sensor and a color filter array (CFA). According to the proposed adaptive heterogeneity projection masks and Sobel- and luminance estimation-based masks, we can extract edge information of each pixel in terms of the direction of variation and the gradient from the mosaic image directly and accurately, and the extracted more accurate edge information will be utilized to assist the design of our proposed new joint demosaicing and zooming algorithm. Based on twenty-four popular testing mosaic images, our proposed new zooming algorithm has better image quality performance in terms of two objective color image quality measures, the color peak signal-to-noise ratio (CPSNR) and the S-CIELAB DeltaE* ab metric, and one subjective color image quality measure, the color artifacts, when compared with several previous zooming algorithms.

Structural information-based image quality assessment using LU factorization

The goal of the objective image quality assessment is to quantitatively measure the image quality of an arbitrary image. The objective image quality measure is desirable if it is close to the subjective image quality assessment such as the mean opinion score. Image quality assessment algorithms are generally classified into two methodologies: perceptual and structural information-based. This paper proposes a structural information-based image quality assessment algorithm, in which LU factorization is used for representation of the structural information of an image. The proposed algorithm performs LU factorization of each of reference and distorted images, from which the distortion map is computed for measuring the quality of the distorted image. Finally, the proposed image quality metric is computed from the two-dimensional distortion map. Experimental results with the laboratory for image and video engineering database images show the efficiency of the proposed method, calibrated by linear and logistic regressions, in terms of the Pearson correlation coefficient and root mean square error. In commercial systems, the proposed algorithm can be used for quality assessment of mobile contents and video coding, which effectively replaces the peak signal to noise ratio or the mean square error.

Combining vector ordering and spatial information for color image interpolation

A new idea for color image zooming techniques is introduced in this paper. Conventional interpolation techniques usually utilize only the spatial information of the neighborhood color pixels around the interpolation positions and do not consider their vectorial characteristics. This paper presents a novel color image interpolation method, which takes both the spatial information and vectorial characteristics into consideration. By combining the spatial distance and the vector-aggregated distance associated with each neighborhood color pixel, the proposed solution first determines the importance of each neighborhood pixel to the current interpolated pixel and then employs a data-adaptive vector filter to estimate the new interpolated pixel. The extensive simulation results demonstrate the validity of the proposed interpolator by yielding excellent visual effect and showing significant performance improvement over the conventional and other vector filtering based interpolation methods in terms of objective image quality measures.

基于方向投影的图像质量评价算法

Objective image quality measure, which is a fundamental and challenging job in image processing, evaluates the image quality consistently with human perception automatically. On the assumption that any image distortion could be modeled as the difference between the directional projection-based maps of reference and distortion images, we propose a new objective quality assessment method based on directional projection for full reference model. Experimental results show that the proposed metrics are well consistent with the subjective quality score.