Whole slide images (WSIs) are fundamental components of modern pathology, aiding pathologists in diagnosing diseases such as cancer. However, artefacts such as blurry regions, folded tissue, or uneven staining, often introduced during biopsy or slide preparation, can affect the accuracy and efficiency of the diagnostic process. Detecting these artefacts is critical to ensure delivering acceptable image quality, guiding pathologists toward diagnostically relevant regions, and improving reliability. In recent years, deep learning has increasingly complemented traditional quality control procedures by enabling rapid detection and localization of such artefacts. This progress, however, depends on the availability of high-quality datasets for training, validation, and testing of the detection models. To support this need, we introduce a comprehensive benchmark dataset for artefact detection obtained from two separate sources, consisting of selected WSIs from The Cancer Genome Atlas (TCGA), as well as WSIs provided by Universitair Ziekenhuis Brussel (UZB). The collected WSIs were subdivided into smaller tiles and annotated by artefact type, making them well-suited for deep learning model development. We hope this dataset serves as a valuable foundation for researchers developing tools to enhance WSI quality and diagnostic accuracy. The dataset is publicly available at: https://gitlab.com/etrovub/interfere/pathodataset25.

The Joint Photographic Experts Group (JPEG) is currently addressing the challenges in assessing plenoptic image quality by developing new standards for both subjective and objective assessment. This process entails revisiting existing recommendations and establishing a visual quality assessment (QA) framework that considers the unique aspects of plenoptic data. The focus is currently on the light field modality, with efforts to gather expert contributions and develop tools that cater both subjective and objective QA, ensuring that the evolving requirements of plenoptic imaging QA are met. This paper presents the JPEG Pleno subjective test tool, a pivotal element in JPEG’s standardization endeavor, designed to facilitate a range of subjective QA experiments that steer the decisions during a standardization process.

Deep features of convolutional neural networks (CNNs), designed for high-level computer vision tasks such as object detection, have been shown to be also effective for image quality assessment (IQA). This motivates further investigations to understand why CNNs are good IQA estimators and, in a broader sense, whether a correspondence exists between CNN's processing stages and the hierarchical mechanism of the human visual system (HVS). In this paper, we stimulate CNNs with a new family of maximally-regular textures to investigate if higher areas of the visual cortex can be approximated within the CNN layers. The results show interesting correspondence between CNNs and the visual cortex, suggesting that these frameworks might be able to serve as suitable baseline candidates for developing a new generation of quality metrics that better replicate the complex stages of the HVS. Developing such IQA models needs in-depth research into the mechanism of the HVS and designing CNNs with better quality-aware feature encoding. As an initial step toward this goal, we established new criteria for improving the performance of pre-trained networks in quality assessment applications by leveraging texture sensitivity. The outcomes illustrate that our feature map weighting and neuron selection criteria could improve the IQA task.

Explainable artificial intelligence (XAI) has gained considerable attention in recent years as it aims to help humans better understand machine learning decisions, making complex black-box systems more trustworthy. Visual explanation algorithms have been designed to generate heatmaps highlighting image regions that a deep neural network focuses on to make decisions. While convolutional neural network (CNN) models typically follow similar processing operations for feature encoding, the emergence of vision transformer (ViT) has introduced a new approach to machine vision decision-making. Therefore, an important question is which architecture provides more human-understandable explanations. This paper examines the explain-ability of deep architectures, including CNN and ViT models under different vision tasks. To this end, we first performed a subjective experiment asking humans to highlight the key visual features in images that helped them to make decisions in two different vision tasks. Next, using the human-annotated images, ground-truth heatmaps were generated that were compared against heatmaps generated by explanation methods for the deep architectures. Moreover, perturbation tests were performed for objective evaluation of the deep models' explanation heatmaps. According to the results, the explanations generated from ViT are deemed more trustworthy than those produced by other CNNs, and as the features of the input image are more dispersed, the advantage of the model becomes more evident.

Opinion-unaware no-reference (OU-NR) methods for image quality assessment (IQA) are of great interest since they can predict visual quality independent of a reference image and knowledge of human quality opinions. Models of image naturalness trained on a corpus of pristine images have shown potential for developing OU-NR methods. However, the extracted features may not match the preferences of the human visual system (HVS). This paper aims to utilize the features of convolutional neural networks to achieve a richer representation of the naturalness space. In addition, the IQA processing steps from training to quality measurement are revisited and the naturalness model is improved by incorporating HVSinspired criteria. Experimental results show the higher performance and generalizability of the naturalness model – constructed using HVS-aligned deep features – under different distortion types and image contents. The source code of the quality index is available at https://gitlab.com/saeedmp/dni.

Objective quality assessment of digital holograms has proven to be a challenging task. While prediction of perceptual quality of the recorded 3D content from the holographic wavefield is an open problem; perceptual quality assessment from content after rendering, requires a time-consuming rendering step and a multitude of possible viewports. In this research, we use 96 Fourier holograms of the recently released HoloDB database to evaluate the performance of well-known and state-of-the-art image quality metrics on digital holograms. We compare the reference holograms with their distorted versions: (i) before rendering on the real and imaginary parts of the quantized complex-wavefield, (ii) after converting Fourier to Fresnel holograms, (iii) after rendering, on the quantized amplitude of the reconstructed data, and (iv) after subsequently removing speckle noise using a Wiener filter. For every experimental track, the quality metric predictions are compared to the Mean Opinion Scores (MOS) gathered on a 2D screen, light field display and a holographic display. Additionally, a statistical analysis of the results and a discussion on the performance of the metrics are presented. The tests demonstrate that while for each test track a few quality metrics present a highly correlated performance compared to the multiple sets of available MOS, none of them demonstrates a consistently high-performance across all four test-tracks.

Human visual system (HVS) strongly responds to motion information. In particular, neurons in middle temporal (MT) area of the brain are sensitive to certain velocities. Thus distortions appeared on moving stimuli, as visually salient information, are visually very important and can significantly affect the quality of experience (QoE). Judder is a visual artifact that manifests as non-smooth motion when tracking a moving object on a digital display. In particular, the effect of judder becomes more significant in the wide field of view (FoV) displays where objects can be observed in a longer trajectory and duration. Head mounted display (HMD) maximizes FoV by enabling full 360° immersive experience and there has been rapid growth in the number of omnidirectional videos for HMDs. However, the impact of judder on visual QoE of the omnidirectional video has never been studied. In this paper, we first established a database of omnidirectional video sequences to study human responses on the visual effect of judder when watching videos on wide FoV HMDs. Two subjective tests were conducted to understand the impact of judder on video sequences compressed at different bit rates (i.e. different picture quality levels). Next, based on the subjective results, we proposed a novel QoE model considering the joint effect of judder, visual masking and picture quality. Experimental results indicate the effectiveness of the proposed model for accurate quality prediction of the videos with judder.

With the emergence of light field (LF) imaging technology, various challenging problems for LF visualization, compression, and transmission have to be addressed. Measuring perceptual quality is of utmost importance to assess the impact of an LF processing step on the visual experience of the finally rendered content to the end-user. In particular, objective quality assessment (QA) plays a key role in the quality optimization of LF imaging systems. In this paper, we conducted a comprehensive experiment to evaluate the performance of different objective QA methods for LF application. To this end, we selected a total number of 250 LFs (more than 48000 perspective views) from three public data sets to evaluate 16 objective QA metrics. Moreover, the subjective scores from three test data sets were aligned to produce an integrated data set for quality evaluation. The performance results across the different data sets aid to choose the most reliable metrics that are consistently performing well under various distortion and content characteristic conditions.

Depth Image-Based Rendering (DIBR) is the key technique in many multi-view 3D applications to synthesize virtual views using texture and depth information. However, DIBR induces distortions that disturb the visual quality of experience; therefore, image quality assessment (IQA) methods are essential to evaluate the quality of the synthesized views. The characteristics of the DIBR-related distortions are different from those of the traditional video coding distortions and conventional objective IQA methods often fail to provide accurate quality predictions for synthesized views. In this letter, we proposed a new Full Reference (FR) objective metric for evaluation of DIBR synthesized views. We used a feature matching method at feature (key) points of reference and synthesized images to quantify the local differences. Moreover, global quality loss is computed in shift-compensated views by measuring the gradient difference in image superpixels. Performance evaluation on the three public data sets shows the effectiveness of the proposed model. A software release of the proposed method is available at https://gitlab.com/etro/ssdi.

High dynamic range (HDR) imaging enables capturing a wide range of luminance levels existing in real-world scenes. While HDR capturing devices become widespread in the market, the display technology is yet limited in representing full luminance ranges and standard low dynamic range (LDR) displays are currently more prevalent. To visualize the HDR content on traditional displays, tone mapping (TM) operators are introduced that convert HDR content into LDR. The dynamic range compression and different processing steps during TM can lead to loss of scene details, as well as luminance and chrominance changes. Such signal deviations will affect image naturalness and consequently disturb the visual quality of experience. Therefore, research into objective methods for quality evaluation of tone-mapped images has received attention in recent years. In this paper, we proposed a completely blind image quality evaluator for tone-mapped images based on a multi-attribute feature extraction scheme. Due to the diversity of TM distortions, various image characteristics are taken into account to develop an effective metric. The features are designed by considering spectral and spatial entropy, detection probability of visual information, image exposure, sharpness, and color properties. The quality-relevant features are then fed into a machine-learning regression framework to pool a quality score. The validation tests on two benchmark datasets reveal the superior performance of the proposed approach compared to the competing metrics.

High Dynamic Range (HDR) imaging captures a wide range of luminance existing in real-world scenes. Due to large luminance levels and higher brightness of HDR displays, artefacts can be more noticeable to the Human Visual System (HVS). In a first attempt to experimentally quantify those noticeable levels for HDR images, we pioneered in conducting an exhaustive and comprehensive Just Noticeable Difference (JND) subjective experiment of which the outcome is presented in this paper. Six distortions including JPEG, JPEG2000, noise, blur, contrast change, and quantization artefacts have been considered in the test. The distortions were applied to 10 HDR images in 100 distortion levels resulting a database of 6000 HDR test images. The subjects were asked to find the image JND location on each set of 100 images they had the freedom to explore. The effect of content features on the noticeable threshold selection is investigated per distortion type. Our results in some cases show a significant correlation between content features and JNDs. We are hoping that our results can contribute to further exploitation of a precise HVS model for HDR quality assessment and optimization of the coding and bit allocation in HDR compression.

During acquisition, processing, compression and transmission, images may be corrupted by multiple distortions such as blur, noise or compression artefacts. However, current image quality assessment (IQA) methods are often designed for images degraded by a single distortion type. This paper proposes a reduced-reference (RR) IQA method to predict the quality of multi-distorted images. The method is based on feature extraction from the reference and the distorted images. Based on internal generative mechanism (IGM) theory, the images are decomposed first into their predicted and disorderly portions. Next, several features are captured from each portion and feature differences are computed between the reference and distorted images. Finally, support vector regression (SVR) is adopted to obtain a quality score. The results on public multiply-distorted image databases, namely MDID2015 and MLIVE, show that the proposed method delivers higher accuracy than several image quality metrics.

Depth from defocus (DFD) technique calculates the blur amount in images considering that the depth and defocus blur are related to each other. Existing blur estimation methods generally compute the blur at edge locations and solve an optimization problem to propagate the blur from edges to all image pixels. Solving the pixel-based optimization problem is time-consuming and it is the performance bottleneck of current approaches. Moreover, the generated depth maps are not consistent in textured areas and the blur estimation may be incorrect in the regions with soft shadows. We address these problems by proposing a superpixel-based blur estimation method. Experimental results show that our superpixel-based method is faster than pixel-based blur estimation and can improve depth data on textured regions and soft shadows.