Quality improvement of adaptive optics retinal images using conditional adversarial networks.

We present a no-reference (NR) image quality assessment (IQA) algorithm that is inspired by the representation of visual scenes in the primary visual cortex of the human visual system. Specifically, we use the sparse coding model of the area V1 to construct an overcomplete dictionary for sparsely representing pristine (undistorted) natural images. First, we empirically demonstrate that the distribution of the sparse representation coefficients of natural images have sharp peaks and heavy tails, and can therefore be modeled using a Univariate Generalized Gaussian Distribution (UGGD). We then show that the UGGD model parameters form good features for distortion estimation and formulate our no-reference IQA algorithm based on this observation. Subsequently, we find UGGD model parameters that are representative of the class of pristine natural images. This is achieved using a training set of undistorted natural images. The perceptual quality of a test image is then defined to be the likelihood of its sparse coefficients being generated from the pristine UGGD model. We show that the proposed algorithm correlates well with subjective evaluation over several standard image databases. Further, the proposed method allows us to construct a distortion map that has several useful applications like distortion localization, adaptive rate allocation etc. Finally and importantly, the proposed NR-IQA algorithm does not make use of any distortion information or subjective scores during the training process.

Prior knowledge about anticipated distortions and their corresponding human opinion scores is needed in the most general purpose no-reference image quality assessment algorithms. When creating the model, all distortion types may not be existed. Predicting the quality of distorted images in practical no-reference image quality assessment algorithms is devised without prior knowledge about images or their distortions. In this study, a blind/no-reference opinion and distortion unaware image quality assessment algorithm based on natural scenes is developed. The proposed approach uses a set of novel features to measure image quality in a spatial domain. The extracted features which are from the scenes gist are formed using Weibull distribution statistics. When testing the proposed algorithm on LIVE database, experiments show that it correlates well with subjective opinion scores. They also show that the proposed algorithm significantly outperforms the popular full-reference peak signal-to-noise ratio (PSNR) and structural similarity (SSIM) methods. Not only do the results reasonably well compete with the recently developed natural image quality evaluator (NIQE) model, but also outperform it.

Objective quality assessment metrics that are consistent with human judgments of image quality, play an important role in many image processing applications. No-reference image quality assessment (NR-IQA) algorithms evaluate the quality of distorted images without any information about the reference images. In this paper, we propose an accurate and simple method for NR-IQA. In this method local binary pattern (LBP) operator is first applied to subbands of wavelet decomposed image separately; then histogram of LBP codes is calculated and concatenated to form quality aware features. Finally, support vector regression (SVR) is utilized to map the feature vectors to subjective quality scores. The experimental results for LIVE and TID2008 databases show that our proposed method is highly correlated to the subjective scores, and competitive to state-of-the-art NR-IQA methods.

Image contrast distortion is a common type of distortion in digital images. However, there is almost no research on the no-reference image quality assessment (NR-IQA) algorithm for image contrast. Therefore, we propose a histogram-based NR-IQA algorithm for contrast distorted images. Firstly, we analyze the image sequence with gradually changing contrast, and the image features are extracted from two aspects: objective statistical attribute, subjective perception attribute. And then we propose three statistical features, including image local contrast, histogram shape and image brightness, which can describe the image quality more simply and intuitively. Furthermore, we introduce the Just noticeable difference (JND) model, which makes the proposed algorithm have a higher matching degree between the human vision system (HVS) and the objective features in the algorithm. Finally, the support vector regression (SVR) is utilized to obtain the mapping relationship between the quantified features and the subjective scores to predict the quality of contrast distorted images. The outstanding performance of the proposed algorithm have been proved on CSIQ, TID2013 and CCID2014 databases.

Existing no-reference image quality assessment (NR-IQA) methods mainly focus on designing the low-level features related to image degradation. However, the evaluation of image quality is the human visual perception of image content, involving the integrated analysis of global high-level semantics and local low-level characteristics. From this perspective, we propose a NR-IQA framework based on global and local content perception. We adopt the deep convolutional neural network (DCNN) to extract the semantic feature implied in global image content. The perception of visual quality associated with local content utilizes the visual attention and filtering mechanisms of human visual system. The overall image quality is estimated by combining the semantic and local characteristic features generated from the perceptions. Experimental results on the LIVE IQA database demonstrate that our method is superior to the state-of-the-art NR-IQA algorithms and competitive to the popular full-reference IQA methods. Further experiments on the TID2008 dataset show that the proposed approach is robust for various kinds of distortion types.

No-reference image quality assessment algorithms: A survey

Measuring visual quality, as perceived by human observers, is becoming increasingly important in many applications where humans are the ultimate consumers of visual information. Significant progress has been made for assessing the subjective quality of natural images, such as those taken by optical cameras. Natural Scene Statistics (NSS) is an important tool for no-reference visual quality assessment of natural images, where the reference image is not needed for comparison. In this paper, we take an important step towards using NSS to automate visual quality assessment of photorealistic synthetic scenes typically found in video games and animated movies. Our primary contributions are (1) conducting subjective tests on our publicly available ESPL Synthetic Image Database containing 500 distorted images (20 distorted images for each of the 25 original images) in 1920 × 1080 format, and (2) evaluating the performance of 17 no-reference image quality assessment (IQA) algorithms using synthetic scene statistics. We find that similar to natural scenes, synthetic scene statistics can be successfully used for IQA and certain statistical features are good for certain image distortions.

BackgroundBrain volume measurement serves as a critical approach for assessing brain health status. Considering the close biological connection between the eyes and brain, this study aims to investigate the feasibility of estimating brain volume through retinal fundus imaging integrated with clinical metadata, and to offer a cost-effective approach for assessing brain health.MethodsBased on clinical information, retinal fundus images, and neuroimaging data derived from a multicenter, population-based cohort study, the KaiLuan Study, we proposed a cross-modal correlation representation (CMCR) network to elucidate the intricate co-degenerative relationships between the eyes and brain for 755 subjects. Specifically, individual clinical information, which has been followed up for as long as 12 years, was encoded as a prompt to enhance the accuracy of brain volume estimation. Independent internal validation and external validation were performed to assess the robustness of the proposed model. Root mean square error (RMSE), peak signal-to-noise ratio (PSNR), and structural similarity index measure (SSIM) metrics were employed to quantitatively evaluate the quality of synthetic brain images derived from retinal imaging data.ResultsThe proposed framework yielded average RMSE, PSNR, and SSIM values of 98.23, 35.78 dB, and 0.64, respectively, which significantly outperformed 5 other methods: multi-channel Variational Autoencoder (mcVAE), Pixel-to-Pixel (Pixel2pixel), transformer-based U-Net (TransUNet), multi-scale transformer network (MT-Net), and residual vision transformer (ResViT). The two- (2D) and three-dimensional (3D) visualization results showed that the shape and texture of the synthetic brain images generated by the proposed method most closely resembled those of actual brain images. Thus, the CMCR framework accurately captured the latent structural correlations between the fundus and the brain. The average difference between predicted and actual brain volumes was 61.36 cm3, with a relative error of 4.54%. When all of the clinical information (including age and sex, daily habits, cardiovascular factors, metabolic factors, and inflammatory factors) was encoded, the difference was decreased to 53.89 cm3, with a relative error of 3.98%. Based on the synthesized brain MR images from retinal fundus images, the volumes of brain tissues could be estimated with high accuracy.ConclusionsThis study provides an innovative, accurate, and cost-effective approach to characterize brain health status through readily accessible retinal fundus images.Trial registration No. NCT05453877 (https://clinicaltrials.gov/).Supplementary InformationThe online version contains supplementary material available at 10.1186/s40779-025-00630-2.

Aim: The objective of this project was to evaluate high resolution images from an adaptive optics retinal imager through comparisons with standard film-based and standard digital fundus imagers. Methods: A clinical prototype adaptive optics fundus imager (AOFI) was used to collect retinal images from subjects with various forms of retinopathy to determine whether improved visibility into the disease could be provided to the clinician. The AOFI achieves low-order correction of aberrations through a closed-loop wavefront sensor and an adaptive optics system. The remaining high-order aberrations are removed by direct deconvolution using the point spread function (PSF) or by blind deconvolution when the PSF is not available. An ophthalmologist compared the AOFI images with standard fundus images and provided a clinical evaluation of all the modalities and processing techniques. All images were also analyzed using a quantitative image quality index. Results: This system has been tested on three human subjects (one normal and two with retinopathy). In the diabetic patient vascular abnormalities were detected with the AOFI that cannot be resolved with the standard fundus camera. Very small features, such as the fine vascular structures on the optic disc and the individual nerve fiber bundles are easily resolved by the AOFI. Conclusion: This project demonstrated that adaptive optic images have great potential in providing clinically significant detail of anatomical and pathological structures to the ophthalmologist.

No-reference image quality assessment in contourlet domain

The automatic analysis of retinal blood vessels plays an important role in the computer-aided diagnosis. In this paper, we introduce a probabilistic tracking-based method for automatic vessel segmentation in retinal images. We take into account vessel edge detection on the whole retinal image and handle different vessel structures. During the tracking process, a Bayesian method with maximum a posteriori (MAP) as criterion is used to detect vessel edge points. Experimental evaluations of the tracking algorithm are performed on real retinal images from three publicly available databases: STARE (Hoover et al., 2000), DRIVE (Staal et al., 2004), and REVIEW (Al-Diri et al., 2008 and 2009). We got high accuracy in vessel segmentation, width measurements, and vessel structure identification. The sensitivity and specificity on STARE are 0.7248 and 0.9666, respectively. On DRIVE, the sensitivity is 0.6522 and the specificity is up to 0.9710.

Improved understanding of the mechanisms underlying inherited retinal degenerations has created the possibility of developing much needed treatments for these relentless, blinding diseases. However, standard clinical indicators of retinal health (such as visual acuity and visual field sensitivity) are insensitive measures of photoreceptor survival. In many retinal degenerations, significant photoreceptor loss must occur before measurable differences in visual function are observed. Thus, there is a recognized need for more sensitive outcome measures to assess therapeutic efficacy as numerous clinical trials are getting underway. Adaptive optics (AO) retinal imaging techniques correct for the monochromatic aberrations of the eye and can be used to provide nearly diffraction-limited images of the retina. Many groups routinely are using AO imaging tools to obtain in vivo images of the rod and cone photoreceptor mosaic, and it now is possible to monitor photoreceptor structure over time with single cell resolution. Highlighting recent work using AO scanning light ophthalmoscopy (AOSLO) across a range of patient populations, we review the development of photoreceptor-based metrics (e.g., density/geometry, reflectivity, and size) as candidate biomarkers. Going forward, there is a need for further development of automated tools and normative databases, with the latter facilitating the comparison of data sets across research groups and devices. Ongoing and future clinical trials for inherited retinal diseases will benefit from the improved resolution and sensitivity that multimodal AO retinal imaging affords to evaluate safety and efficacy of emerging therapies.

Adaptive optics full-filed OCT (FFOCT) with a transmissive liquid crystal spatial light modulator (LCSLM) as wavefront corrector is used without strict plane conjugation for low order aberrations corrections. We validated experimentally that FFOCT resolution is independent of aberrations and only reduce the signal level. A signal based sensorless algorithm was thus applied for wavefront distortion compensation. Image quality improvements by the wavefront sensorless control of the LCSLM were evaluated on in vitro samples. By replacing the FFOCT sample arm objective with an artificial eye used to train ophthalmologists, adaptive optics retinal imaging was achieved. In vivo experiments using a liquid lens to correct focus and astigmatism are underway.

No Reference Image Quality Assessment (NR-IQA) refers to algorithms that predict the quality of distorted image where the reference image is not available. NR-IQA algorithms are divided into two categories: specific distortion quality assessment and general purpose quality assessment. The first type of algorithms deals with a specific distortion and predict the quality of the image based on this distortion. Which means it assumes that the distortion in the image is known. On the other hand, the general purpose quality assessment type predicts the quality of image with no information about the distortion that affecting the tested image. A two-stage framework, is proposed by Moorthy and Bovik [1], which classifies the distortion followed by distortion-specific quality assessment method. Our proposal here is to improve the classification portion by investigating the performance of different classification techniques and different features. Each feature is validated using different features evaluation techniques. As a result, we construct a set of optimal features that classify image distortions with a high accuracy rate.

This paper proposes image quality measures based on multifractality preservation. Image quality measures play critical roles in designing and evaluating image processing schemes and performance, especially in which the resulting images deviate from the original ones. Examples of such schemes include image denoising and lossy compression. Traditional quality or distortion measures have been based on mean square error (MSE) measure or its derivatives, e.g., signal to noise ratio (SNR) or peak SNR (PSNR). Although such measures often lead to optimal schemes (with respect to MSE or PSNR), they are known to remove image parts that has noise-like appearances. Furthermore, they treat image singularities such as sharp edges or high textures (that are more important visually and diagnostically) and other image parts (that are less important) uniformly. In contrast, multifractal measures proposed in this paper characterize image singularities. This means the measures pay attention more on important image features, such as sharp edges and high textures. Furthermore, it can distinguish noise from noise-like signals through their differences in their types of singularities. As a result, the measure can be used to assess image quality of sensitive images resulting from processing schemes. The paper shows various ways of defining the measure that reveals multifractality of different aspects of images. It reports the use of the multifractal measure to compare a joint photographic expert group (JPEG) scheme and a variant of differential pulse code modulation (DPCM) coding showing that the DPCM scheme is superior in multifractal preservation for comparable compression ratios. As a result, the DPCM coding has been selected for a database of aerial ortho images.