Exploiting neural models for no-reference image quality assessment
20
Citation
15
Reference
10
Related Paper
Citation Trend
Abstract:
We propose an improved algorithm for no-reference image quality assessment (NR-IQA) using the convolutional neural network (CNN) and neural theory based saliency detection. Firstly, we extract non-overlapping patches from the input image. For each patch, we obtain the quality score by CNN network, which consists of seven layers and integrates feature learning and regression into image patch quality estimation. Considering that the patches attracting much attention take significant role in visual perception, an efficient technique based on free energy based neural model is used to detect the saliency map. This saliency map is then applied as a weighting mask to output the quality score of the whole image. Results of experiments show that our algorithm achieves state-of-the-art performance, as compared with the prevailing IQA methods.Keywords:
Feature (linguistics)
Quality Score
Abstract Background Generative Adversarial Networks (GANs) can synthesize brain images from image or noise input. So far, the gold standard for assessing the quality of the generated images has been human expert ratings. However, due to limitations of human assessment in terms of cost, scalability, and the limited sensitivity of the human eye to more subtle statistical relationships, a more automated approach towards evaluating GANs is required. New method We investigated to what extent visual quality can be assessed using image quality metrics and we used group analysis and spatial independent components analysis to verify that the GAN reproduces multivariate statistical relationships found in real data. Reference human data was obtained by recruiting neuroimaging experts to assess real Magnetic Resonance (MR) images and images generated by a Wasserstein GAN. Image quality was manipulated by exporting images at different stages of GAN training. Results : Experts were sensitive to changes in image quality as evidenced by ratings and reaction times, and the generated images reproduced group effects (age, gender) and spatial correlations moderately well. We also surveyed a number of image quality metrics which consistently failed to fully reproduce human data. While the metrics Structural Similarity Index Measure (SSIM) and Naturalness Image Quality Evaluator (NIQE) showed good overall agreement with human assessment for lower-quality images (i.e. images from early stages of GAN training), only a Deep Quality Assessment (QA) model trained on human ratings was sensitive to the subtle differences between higher-quality images. Conclusions We recommend a combination of group analyses, spatial correlation analyses, and both distortion metrics (SSIM, NIQE) and perceptual models (Deep QA) for a comprehensive evaluation and comparison of brain images produced by GANs.
Quality Score
Mean opinion score
Similarity (geometry)
Quality Assessment
Cite
Citations (0)
Image Quality Assessment (IQA) is widely used in digital image processing, and No Reference IQA (NR-IQA) has become the research focus recently. This paper proposes an NR-IQA method based on local structure, which chooses strong structure areas by using local gradients, and assesses the quality of image by utilizing the Maximum Local Gradients (MLG) of strong structure areas. The main novelties are: pixel , s quality assessment based on MLG; whole image quality based on strong edge points , quality. The proposed method can assess noise image and blur image at the same time, and the score of the proposed method is smaller when the distortion is more serious. The results show that the proposed no-reference method for the quality prediction of noise and blur images has a comparable performance to the leading metrics available in literature.
Distortion (music)
Quality Score
Cite
Citations (8)
We propose a family of image quality assessment (IQA) models based on natural scene statistics (NSS), that can predict the subjective quality of a distorted image without reference to a corresponding distortionless image, and without any training results on human opinion scores of distorted images. These `completely blind' models compete well with standard non-blind image quality indices in terms of subjective predictive performance when tested on the large publicly available `LIVE' Image Quality database.
Mean opinion score
Quality Score
Quality Assessment
Cite
Citations (10)
1415 Objectives: Research in nuclear medicine digital image processing requires processed images to be evaluated by nuclear medicine physicians (NMPs). However, it is difficult to get the time of NMPs for evaluating huge number of processed images. Perceptual image quality evaluator (PIQE score) is a no reference image quality evaluator. In this study, we have evaluated the performance of PIQE score for 99mTc-MDP bone scan images.
Materials and Methods: Forty two 99mTc-MDP bone scan studies i.e. 84 images were restored using Richardson-Lucy algorithm. During the process of restoration, for each input image, 60 processed images were generated using different parameters of algorithm. Thus, the process generated 5124 images (61x84). For 5 studies (10 images), input and processed images were compared by a NMP to decide one combination of parameters that resulted in high quality processed image. Then for all 42 studies the image with the selected parameters was picked which yielded image series “A”. Based on PIQE score evaluator high quality images from each study were selected and image series “B” was formed. The quality of Image in series A and B were compared with corresponding input images by two NMPs. They were asked to score the quality of image by assigning one of the six numeric score with consensus (0, 1, 2, 3, 4 and 5), where score “0” is for very bad and annoying enhancement (the image quality is totally distorted), score “3” is for no noticeable enhancement (natural and similar to original image), score 5 is for significant enhancement without annoying distortions (looks natural across the overall image) and other values were selected according to the perceived image quality.
Results: In 61.90% of the cases (52 out of 84 images) the high quality image selected based on PIQE score and high quality image selected based on NMPs evaluation were concordant. The difference in image quality score between two image series was equal 2 in only 5.97% of cases (5 out of 84 images). There were 28.57% images (24 images) in which the difference in image quality score was equal to 1. Visually, images in which the difference in image quality score was equal to 1; both images had appreciable enhancement with slight difference in degree of enhancement in some regions which was acceptable to NMPs. If we combine the images with same and acceptable difference of 1 in image quality score together, then overall 90.47% ( 76 out of 84) of the images selected based on PIQE score were acceptable to the NMPs.
Conclusions: Our results demonstrate that although there is no replacement of visual image quality assessment made by NMPs however, PIQE score may be used during the initial part of the research as an image quality evaluator.
Quality Score
Cite
Citations (5)
Quantifying image quality through subjective evaluation is very critical to image quality evaluation. Using the image quality ruler method, an average score per stimulus can be easily obtained in the unit of Just Noticeable Differences (JNDs). However, it requires a large number of subjects, since pure averaging does not consider the different judging quality of different subjects. In this paper, we propose an image quality evaluation framework using the image quality ruler method with a statistical model. By incorporating this model, we consider the quality score, the expertise of the subjects, and the difficulty of image rating task as three hidden variables. Then we use expectation-maximization (EM) to estimate these hidden variables. From our experimental results, we show that our method provides reliable results without using a large number of subjects. Preliminary results also demonstrate that the estimates of the parameters can guide us to better distribute the valuable human resources used to conduct psychophysical experiments.
Ruler
Quality Score
Maximization
Cite
Citations (5)
Measurement of visual quality is of fundamental importance for numerous image and video processing applications, where the goal of quality assessment (QA) algorithms is to automatically assess the quality of images or videos in agreement with human quality judgments. Over the years, many researchers have taken different approaches to the problem and have contributed significant research in this area and claim to have made progress in their respective domains. It is important to evaluate the performance of these algorithms in a comparative setting and analyze the strengths and weaknesses of these methods. In this paper, we present results of an extensive subjective quality assessment study in which a total of 779 distorted images were evaluated by about two dozen human subjects. The "ground truth" image quality data obtained from about 25,000 individual human quality judgments is used to evaluate the performance of several prominent full-reference image quality assessment algorithms. To the best of our knowledge, apart from video quality studies conducted by the Video Quality Experts Group, the study presented in this paper is the largest subjective image quality study in the literature in terms of number of images, distortion types, and number of human judgments per image. Moreover, we have made the data from the study freely available to the research community. This would allow other researchers to easily report comparative results in the future.
Ground truth
Quality Score
Distortion (music)
Strengths and weaknesses
Quality Assessment
Cite
Citations (2,709)
Determining the perceived quality of a digital image can be done by showing it to a large group of people and ask them to rate its quality. However, this method is too cumbersome and time-consuming for most applications. Therefore, automatic metrics have been developed, which are able to objectively predict the perceived image quality without the need for any human input. Such image quality assessment metrics average the predicted quality across a whole image into a single quality value. To improve their performance, their predicted image quality is weighted with visual attention information: regions in the image that receive more attention are weighted more heavily in the quality assessment. Furthermore, the effect of a quality assessment task on visual attention is investigated via a large scale subjective experiment. The main findings are that 1) people who are looking freely pay more attention to the region of interest than people who are scoring the image quality, 2) applying the visual attention of people who are looking freely to image quality assessment metrics yields a higher performance gain than the visual attention of people who are scoring the image quality, and 3) the predicted image quality in the region of interest has a more positive influence on the overall predicted image quality than the quality in the background. In short, visual attention information can be used to increase the performance of image quality assessment metrics.
Quality Score
Cite
Citations (0)
Quality Assessment
Cite
Citations (18)
Traditionally, image quality estimators have been designed and optimized to operate over the entire quality range of images in a database, from very low quality to visually lossless. However, if quality estimation is limited to a smaller quality range, their performances drop dramatically, and many image applications only operate over such a smaller range. This paper is concerned with one such range, the low-quality regime, which is defined as the interval of perceived quality scores where there exists a linear relationship between the perceived quality scores and the perceived utility scores and exists at the low-quality end of image databases. Using this definition, this paper describes a subjective experiment to determine the low-quality regime for databases of distorted images that include perceived quality scores but not perceived utility scores, such as CSIQ and LIVE. The performances of several image utility and quality estimators are evaluated in the low-quality regime, indicating that utility estimators can be successfully applied to estimate perceived quality in this regime. Omission of the lowestfrequency image content is shown to be crucial to the performances of both kinds of estimators. Additionally, this paper establishes an upper-bound for the performances of quality estimators in the LQR, using a family of quality estimators based on VIF. The resulting optimal quality estimator indicates that estimating quality in the low-quality regime is robust to exact frequency pooling weights, and that near-optimal performance can be achieved by a variety of estimators providing that they substantially emphasize the appropriate frequency content.
Pooling
Quality Score
Cite
Citations (2)