dipIQ: Blind Image Quality Assessment by Learning-to-Rank Discriminable Image Pairs

By Kede Ma, Wentao Liu, Tongliang Liu, Zhou Wang, Dacheng Tao. IEEE Transactions on Image Processing, 2017

339 citations recorded by the source corpus.

Objective assessment of image quality is fundamentally important in many image processing tasks. In this paper, we focus on learning blind image quality assessment (BIQA) models, which predict the quality of a digital image with no access to its original pristine-quality counterpart as reference. One of the biggest challenges in learning BIQA models is the conflict between the gigantic image space (which is in the dimension of the number of image pixels) and the extremely limited reliable ground truth data for training. Such data are typically collected via subjective testing, which is cumbersome, slow, and expensive. Here, we first show that a vast amount of reliable training data in the form of quality-discriminable image pairs (DIPs) can be obtained automatically at low cost by exploiting large-scale databases with diverse image content. We then learn an opinion-unaware BIQA (OU-BIQA, meaning that no subjective opinions are used for training) model using RankNet, a pairwise learning-to-rank (L2R) algorithm, from millions of DIPs, each associated with a perceptual uncertainty level, leading to a DIP inferred quality (dipIQ) index. Extensive experiments on four benchmark IQA databases demonstrate that dipIQ outperforms the state-of-the-art OU-BIQA models. The robustness of dipIQ is also significantly improved as confirmed by the group MAximum Differentiation competition method. Furthermore, we extend the proposed framework by learning models with ListNet (a listwise L2R algorithm) on quality-discriminable image lists (DIL). The resulting DIL inferred quality index achieves an additional performance gain.

Subfields: Computer Vision and Pattern Recognition

View Paper: https://www.semanticscholar.org/paper/551286b29d06865f45ef2da98e6ad5be8e9d2c28

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. Wangmeng Zuo
2. Kede Ma
3. Luc Van Gool
4. Xiaokang Yang
5. Kai Zhang
6. Lei Zhang
7. Radu Timofte
8. Dacheng Tao

Citations in the corpus

• Blind Image Quality Estimation via Distortion Aggravation (2018, background): “ture learning [29], [30] and rank learning [31].”
• Blind Image Quality Assessment Using a Deep Bilinear Convolutional Neural Network (2019, methodology): “1) Performance on Individual Databases: We compare the proposed model against several state-of-the-art BIQA methods: BRISQUE [7], M3 [48], FRIQUEE [22], CORNIA [8], HOSA [49], and dipIQ [27], whose source codes are provided by the respective authors.”
• Deep CNN-Based Blind Image Quality Predictor (2019, methodology): “ating discriminable image pairs [22], or employing multitask learning [23].”
• Blindly Assess Image Quality in the Wild Guided by a Self-Adaptive Hyper Network (2020, background): “Noticing the limited size of training data from existing IQA databases, [24] and [26] proposed to generate vast training samples by labeling their quality rank instead of quality score.”
• No-Reference Image Quality Assessment via Transformers, Relative Ranking, and Self-Consistency (2021, background): “Before the raise of deep learning, early version of general-purpose NR-IQA methods mainly divide into natural scene statistics (NSS) based metrics [16, 17, 6, 5, 18, 19, 8, 20, 21] and learning-based metrics [22, 23, 24, 25, 26, 27].”
• End-to-End Blind Image Quality Assessment Using Deep Neural Networks (2018, methodology): “The competing algorithms are chosen to cover a diversity of design philosophies, including three classic ones: DIIVINE [3], BRISQUE [4] and CORNIA [38], and five state-of-the-art ones: ILNIQE [40], BLISS [22], HOSA [41], dipIQ [23] and”
• Perceptual Quality Assessment of Smartphone Photography (2020, background): “Early BIQA models [4, 20–23, 39 ] mainly focus on synthetic distortions,”
• Uncertainty-Aware Blind Image Quality Assessment in the Laboratory and Wild (2020, methodology): “obtained from distortion specifications [3], [14], [30], fullreference IQA models [15], [16], and human data [17].”