Special Issue on Recent Advances on Deep Learning for Media Quality Modeling
摘要截稿:
全文截稿: 2019-04-01
影响因子: 2.779
期刊难度:
CCF分类: C类
中科院JCR分区:
• 大类 : 工程技术 - 3区
• 小类 : 工程:电子与电气 - 3区
Overview
Media quality assessment means evaluate image/video quality subjectively or objectively. Subjective evaluation means identifying whether a photo or video clip is aesthetically pleasing. Meanwhile, objective evaluation means identifying image/video distortion, noise level, etc. In the past decades, researchers have proposed a variety of shallow computational models, aiming at mimicking the process of assessing media quality. Owing to the remarkable progress made in deep-learning-based image recognition systems, an increasing number of deep architectures are proposed to hierarchically characterize media quality. Compared to the domain knowledge required for engineering shallow quality-related features, deep quality models are typically trained in an end-to-end and black-box setting. Empirical results have demonstrated the competitive performance of the deep quality features. Despite these advantages, current deep models are still far from satisfactory due to the following limitations. First, how to deeply encode the complicated quality-related features cues into the current deep quality models is difficult. Second, the spatial interactions among regions can arouse human perception of visual subjective quality, but existing deep models cannot explicitly capture such feature. Third, the state-of-the-art quality datasets, such as the AVA and MICT, are intolerably small for training a deep model directly. How to effectively train a deep neural network based on a small number of samples is an urgent problem. Fourth, how to systematically develop datasets, benchmarks and evaluation platforms to test the performance of those deep quality models.
This special issue will focus on the most recent(especially the past five years)progress in utilizing novel deep architectures to model image/video subjective/object quality. We target the outstanding work of upgrading the state-of-the-art deep quality model in order to effectively capture locally distributed details. Submissions related to new quality benchmarks for testing the performance of deep learning models are also welcome. The primary objective of this special issue is to foster focused attention on the latest research progress in this cutting-edge area. We solicit original contributions that address the challenges of media quality modeling with various deeply-learned features. This special issue targets researchers and practitioners from both industry and academia.
The topics of interest include (but are not limited to):
New deep computational models for media quality evaluation;
Deep quality models for enhance the shallow-feature-based intelligent systems;
Quality-driven video/image processing techniques;
Semantic models for deep multimedia quality prediction;
Multi-view learning algorithms for deep media quality modeling;
Visual aesthetic prediction for photo and video management systems;
Leveraging human interactions to improve deep quality models;
Data-driven deep quality models for Internet-scale media retrieval;
Novel deep quality features and their applications in pattern recognition.
Deep models trained using small-scale samples for visual quality prediction;
Novel photo or video retargeting/cropping/re-composition using deep features;
Datasets, benchmarks, and validation of deep quality models;
Subjective methodologies to estimate the quality in real-world systems;
Novel data mining techniques for media quality analysis, including multimodal, affective, and brain-computer interfaces;
Novel visualization technologies for deep quality features;