Efficient and effective objective image quality assessment metrics
Bibliographic record
Abstract
Acquisition, transmission, and storage of images and videos have been largely increased in recent years. At the same time, there has been an increasing demand for high quality images and videos to provide satisfactory quality-of-experience for viewers. In this respect, high dynamic range (HDR) imaging with higher than 8-bit depth has been an interesting approach in order to capture more realistic images and videos. Objective image and video quality assessment plays a significant role in monitoring and enhancing the image and video quality in several applications such as image acquisition, image compression, multimedia streaming, image restoration, image enhancement and displaying. The main contributions of this work are to propose efficient features and similarity maps that can be used to design perceptually consistent image quality assessment tools. In this thesis, perceptually consistent full-reference image quality assessment (FR-IQA) metrics are proposed to assess the quality of natural, synthetic, photo-retouched and tone-mapped images. In addition, efficient no-reference image quality metrics are proposed to assess JPEG compressed and contrast distorted images. Finally, we propose a perceptually consistent color to gray conversion method, perform a subjective rating and evaluate existing color to gray assessment metrics. \n \nExisting FR-IQA metrics may have the following limitations. First, their performance is not consistent for different distortions and datasets. Second, better performing metrics usually have high complexity. We propose in this thesis an efficient and reliable full-reference image quality evaluator based on new gradient and color similarities. We derive a general deviation pooling formulation and use it to compute a final quality score from the similarity maps. Extensive experimental results verify high accuracy and consistent performance of the proposed metric on natural, synthetic and photo retouched datasets as well as its low complexity. \n \nIn order to visualize HDR images on standard low dynamic range (LDR) displays, tone-mapping operators are used in order to convert HDR into LDR. Given different depth bits of HDR and LDR, traditional FR-IQA metrics are not able to assess the quality of tone-mapped images. The existing full-reference metric for tone-mapped images called TMQI converts both HDR and LDR to an intermediate color space and measure their similarity in the spatial domain. We propose in this thesis a feature similarity full-reference metric in which local phase of HDR is compared with the local phase of LDR. Phase is an important information of images and previous studies have shown that human visual system responds strongly to points in an image where the phase information is ordered. Experimental results on two available datasets show the very promising performance of the proposed metric. \n \nNo-reference image quality assessment (NR-IQA) metrics are of high interest because in the most present and emerging practical real-world applications, the reference signals are not available. In this thesis, we propose two perceptually consistent distortion-specific NR-IQA metrics for JPEG compressed and contrast distorted images. Based on edge statistics of JPEG compressed images, an efficient NR-IQA metric for blockiness artifact is proposed which is robust to block size and misalignment. Then, we consider the quality assessment of contrast distorted images which is a common distortion. Higher orders of Minkowski distance and power transformation are used to train a low complexity model that is able to assess contrast distortion with high accuracy. For the first time, the proposed model is used to classify the type of contrast distortions which is very useful additional information for image contrast enhancement. \n \nUnlike its traditional use in the assessment of distortions, objective IQA can be used in other applications. Examples are the quality assessment of image fusion, color to gray image conversion, inpainting, background subtraction, etc. In the last part of this thesis, a real-time and perceptually consistent color to gray image conversion methodology is proposed. The proposed correlation-based method and state-of-the-art methods are compared by subjective and objective evaluation. Then, a conclusion is made on the choice of the objective quality assessment metric for the color to gray image conversion. The conducted subjective ratings can be used in the development process of quality assessment metrics for the color to gray image conversion and to test their performance.
Fetched live from OpenAlex and de-inverted. Abstracts are not stored in this database: the inverted indexes are 8.6 GB of the frame’s 9.3 GB of text, and the host has 13 GB free.
How this classification was reachedexpand
Full frame machine prediction
Teacher imitationNot calibrated prevalence, not ground truth. Human validation pending. The Gemma side is a direct model label for every work in the frame, read from the title-only record. The Codex side is a classifier learned from the 10,348 direct Codex labels and calibrated to design-weighted sample rates; fields without enough sample support carry no Codex call. Candidate is the union of the two sides; consensus is their intersection. These outputs are machine_predicted_unvalidated and are not human labels.
Distilled classifier scores by category (both heads)
| Category | Codex | Gemma |
|---|---|---|
| Metaresearch | 0.003 | 0.010 |
| Meta-epidemiology (narrow) | 0.002 | 0.000 |
| Meta-epidemiology (broad) | 0.001 | 0.001 |
| Bibliometrics | 0.003 | 0.002 |
| Science and technology studies | 0.000 | 0.001 |
| Scholarly communication | 0.003 | 0.003 |
| Open science | 0.001 | 0.001 |
| Research integrity | 0.001 | 0.001 |
| Insufficient payload (model declined to judge) | 0.002 | 0.001 |
Machine scores (provisional)
The two teacher heads of the student model, read on this work. A score orders the frame for review; it never asserts a category, and the validation status ships verbatim with every row.
Baseline scores from an immature model (maturity gate not passed, 7 training rounds). Scores rank; they never assert a category.
score_only:v0-immature-baseline · verbatim from the scoring run: score_only means the number may rank works, and no category label ships from itClassification
machine, unvalidatedMachine predicted; a candidate call from one source (direct Gemma or distilled Codex), not a consensus.
How this classification was reached, model by model and score by score, is at the end of the page under "How this classification was reached".