Blind Image Quality Assessment via Deep Learning

By Weilong Hou, Xinbo Gao, Dacheng Tao, Xuelong Li. IEEE Transactions on Neural Networks and Learning Systems, 2015

373 citations recorded by the source corpus.

This paper investigates how to blindly evaluate the visual quality of an image by learning rules from linguistic descriptions. Extensive psychological evidence shows that humans prefer to conduct evaluations qualitatively rather than numerically. The qualitative evaluations are then converted into the numerical scores to fairly benchmark objective image quality assessment (IQA) metrics. Recently, lots of learning-based IQA models are proposed by analyzing the mapping from the images to numerical ratings. However, the learnt mapping can hardly be accurate enough because some information has been lost in such an irreversible conversion from the linguistic descriptions to numerical scores. In this paper, we propose a blind IQA model, which learns qualitative evaluations directly and outputs numerical scores for general utilization and fair comparison. Images are represented by natural scene statistics features. A discriminative deep model is trained to classify the features into five grades, corresponding to five explicit mental concepts, i.e., excellent, good, fair, poor, and bad. A newly designed quality pooling is then applied to convert the qualitative labels into scores. The classification framework is not only much more natural than the regression-based models, but also robust to the small sample size problem. Thorough experiments are conducted on popular databases to verify the model's effectiveness, efficiency, and robustness.

Subfields: Computer Vision and Pattern Recognition, Media Technology

View Paper: https://www.semanticscholar.org/paper/72808bcd6e05de33261fc27bcafdb4d1855daf24

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

In Computer Vision and Pattern Recognition

1. Dacheng Tao
2. Junwei Han
3. Kede Ma
4. Gong Cheng
5. Ling Shao
6. Ming-Hsuan Yang
7. Jongyoo Kim
8. Luc Van Gool

In Media Technology

1. Dacheng Tao
2. Kede Ma
3. Jongyoo Kim
4. Sanghoon Lee
5. Weisi Lin
6. Ke Gu
7. Junwei Han
8. Xuelong Li

Citations in the corpus

• Learning Student Networks via Feature Embedding (2018, background): “DEEP neural networks (DNNs) have provided state-of-theart performance in various fields, such as image classification [1], [2], semantic modeling [3], [4], visual quality evaluation [5], object detection [6], [7], and segmentation [8].”
• Deep Learning of Human Visual Sensitivity in Image Quality Assessment Framework (2017, methodology): “marked: SESANIA [7], CNN [8], a ‘Patchwise’ method in [2], and BIECON [9].”
• Learning a No-Reference Quality Assessment Model of Enhanced Images With Big Data (2018, methodology): “Differing from current works which just use a small number of training data [17], [18], [25], [26], [28], [36], we have gathered beyond 100 000 enhanced images (much”
• Broad Learning System: An Effective and Efficient Incremental Learning System Without the Need for Deep Architecture (2018, background): “DEEP structure neural networks and learnings have been applied in many fields and have achieved breakthrough successes in a great number of applications [1], [2], particularly in large scale data processing [3], [4].”
• On the use of deep learning for blind image quality assessment (2016, methodology): “E [12,14] 0.95 0.93 ShearletIQM [29] 0.94 0.93 MGMSD [1] 0.97 0.97 Low Level Features [21] 0.95 0.94 Rectier Neural Network [45] { 0.96 Multi-task CNN [20] 0.95 0.95 Shallow CNN [19] 0.95 0.96 DLIQA [16] 0.93 0.93 HOSA [49] 0.95 0.95 CNN-Prewitt [27] 0.97 0.96 CNN-SVR [26] 0.97 0.96 DeepBIQ 0.98 0.9”
• TOPIQ: A Top-Down Approach From Semantics to Distortions for Image Quality Assessment (2023, background): “They are generally improved with more advanced network architecture, from deep belief net [31] to CNN [32], then to deeper CNN [33]–[35], later to ResNet [36]–[38], and now vision transformers [39]–[41].”
• Lightweight Pyramid Networks for Image Deraining (2018, methodology): “Learning-based methods and convolutional neural networks (CNN) have proven useful for a variety of high-level vision tasks [16]–[23], as well as various image processing problems [24]–[28].”
• Deep CNN-Based Blind Image Quality Predictor (2019, methodology): “Deep learning has lately been adopted in a few NR-IQA studies as a different method from conventional approaches based on NSS [3], [4].”
• dipIQ: Blind Image Quality Assessment by Learning-to-Rank Discriminable Image Pairs (2017, background): “pre-train a model without labels and to easily scale up have also been exploited for this purpose [40], [47], [51], [62].”
• Fully Deep Blind Image Quality Predictor (2017, methodology): “Hou et al. adopted a deep belief network (DBN)-based framework, where the NSS-related features were extracted in the wavelet domain [37].”
• Two-Stage Learning to Predict Human Eye Fixations via SDAEs (2016, background): “recognition [40], [41], the entire deep network can be trained end-to-end, which integrates the feature extraction and classifier establishment into a unified hierarchical model.”