Generating Image Distortion Maps Using Convolutional Autoencoders With Application to No Reference Image Quality Assessment

Dendi, Sathya Veera Reddy; Dev, Chander; Kothari, Narayan; Channappayya, Sumohana S.

doi:10.1109/lsp.2018.2879518

Cited by 42 publications

(17 citation statements)

References 23 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Dendi et al [44] used a convolutional auto-encoder for distortion map generation. The training ground-truth distortion map was estimated by a well-known FR-IQA measure, i.e., SSIM.…”

Section: Conventional Biqa Methodsmentioning

confidence: 99%

“…With the help of "deeper" architectures, the performance of DeepBIQ (InceptionResNetV2) increased by more than 0.1 compared to the best conventional BIQA methods. By training and testing on entire images to preserve content information, Kon- [44] 0.700 0.710 --Learning-to-Rank IQA [45] 0.892 ---KonCept512 0.921 0.937 0.825 0.848 * To reduce computational cost, we reduced CORNIA and HOSA feature vector from 20,000 dimensions and 14,700 dimensions respectively to 100 dimensions using PCA.…”

Section: Comparison With State-of-the-art Biqa Methodsmentioning

confidence: 99%

See 1 more Smart Citation

KonIQ-10k: An ecologically valid database for deep learning of blind image quality assessment

Hosu,

Lin,

Sziranyi

et al. 2019

Preprint

View full text Add to dashboard Cite

Deep learning methods for image quality assessment (IQA) are limited due to the small size of existing datasets. Extensive datasets require substantial resources both for generating publishable content, and annotating it accurately. We present a systematic and scalable approach to create KonIQ-10k, the largest IQA dataset to date consisting of 10,073 quality scored images. This is the first in-the-wild database aiming for ecological validity, with regard to the authenticity of distortions, the diversity of content, and quality-related indicators. Through the use of crowdsourcing, we obtained 1.2 million reliable quality ratings from 1,459 crowd workers, paving the way for more general IQA models. We propose a novel, deep learning model (KonCept512), to show an excellent generalization beyond the test set (0.921 SROCC), to the current state-of-the-art database LIVE-in-the-Wild (0.825 SROCC). The model derives its core performance from the InceptionResNet architecture, being trained at a higher resolution than previous models (512 × 384). A correlation analysis shows that KonCept512 performs similar to having 9 subjective scores for each test image.

show abstract

“…Dendi et al [44] used a convolutional auto-encoder for distortion map generation. The training ground-truth distortion map was estimated by a well-known FR-IQA measure, i.e., SSIM.…”

Section: Conventional Biqa Methodsmentioning

confidence: 99%

Section: Comparison With State-of-the-art Biqa Methodsmentioning

confidence: 99%

KonIQ-10k: An ecologically valid database for deep learning of blind image quality assessment

Hosu,

Lin,

Sziranyi

et al. 2019

Preprint

View full text Add to dashboard Cite

show abstract

“…Structural Similarity Index metric (SSIM) [17] and Semantic Interpretability Score (SIS) [18] are the other two metrics, which are commonly used to evaluate the image quality. For SSIM, there has a similarity between images and the real objects, which is determined by ( 7):…”

Section: Simulation Under Various Incident Beamsmentioning

confidence: 99%

Influence of Beam Distribution on the Quality of Compressed Sensing-Based THz Imaging

Zhang

Zhu

et al. 2020

IEEE Access

View full text Add to dashboard Cite

This paper investigates the impact of incident beam inhomogeneity on the quality of THz imaging based on the compressed sensing (CS) method. Image sampling and reconstructions under point and Gaussian beams with various geometric parameters are compared with the standard one. The simulation results show that the geometric parameters of beams strongly affect the peak signal-to-noise ratio (PSNR) of the reconstructed images. Especially for the Gaussian one, expanding the beam size at the position of the mask (BZPM) dramatically increases the PSNR. To achieve high sensitivity and resolution, new measurement matrices correlating to the incident beam distribution are proposed and the simulation results are demonstrated. Experiments under VDI THz source reveal that by using the new matrices, the PSNR of CS-based imaging at 100 GHz is evidently improved from 6 dB to 13 dB, informing the new measurement matrices are highly efficient and accurate in removing the beam effect on CS-based THz imaging. Our results may provide a new way for the high-quality CS-based THz imaging for target recognition.INDEX TERMS terahertz imaging, compressed sensing, beam inhomogeneity, beam noise removal

show abstract

“…Secondly, it is usually difficult to propose a novel network architecture that could be powerful on both synthetic and authentic distortions. For the first obstacle, to avoid the overfitting problem induced by the insufficient IQA databases, many CNNbased methods [17][18][19][20] crop large images into sub-images for data augmentation. Besides, the random horizontal flip method and transfer learning method are further employed in this paper.…”

Section: Introductionmentioning

confidence: 99%

No‐reference image quality assessment based on multiscale feature representation

Qiao

Zhao

et al. 2021

IET Image Processing

View full text Add to dashboard Cite

The no-reference image quality assessment (NR-IQA) method can evaluate the distortions in an image without the reference image. However, due to the diversity of the image contents and distortion types, it is hard for the existing NR-IQA algorithms to obtain competitive performance on both synthetically and authentically distorted images. To address the problem, a multiscale feature representation-based NR-IQA method that performs well for both synthetic and authentic distortions is proposed. This model consists of two parts: The feature extraction part and the feature fusion part. First, part of the Res2Net-50 network is chosen as the feature extraction part due to its high ability in increasing the range of receptive fields. Then, the feature fusion part consisting of a novel residual block and two fully connected layers is designed to fuse the extracted features and realize the quality score mapping. After a series of stepwise optimization experiments, the most competitive network architecture consisting of the feature extraction part and the feature fusion part is obtained. Comprehensive experiments on the LIVE, TID2013, CSIQ, KADID-10k, KonIQ-10K, and LIVE challenge databases demonstrate that the proposed method can work powerfully on both the synthetic and authentic distortions and also has a strong generalization ability.This is an open access article under the terms of the Creative Commons Attribution-NonCommercial-NoDerivs License, which permits use and distribution in any medium, provided the original work is properly cited, the use is non-commercial and no modifications or adaptations are made.

show abstract

Generating Image Distortion Maps Using Convolutional Autoencoders With Application to No Reference Image Quality Assessment

Cited by 42 publications

References 23 publications

KonIQ-10k: An ecologically valid database for deep learning of blind image quality assessment

KonIQ-10k: An ecologically valid database for deep learning of blind image quality assessment

Influence of Beam Distribution on the Quality of Compressed Sensing-Based THz Imaging

No‐reference image quality assessment based on multiscale feature representation

Contact Info

Product

Resources

About