MPDNet: An underwater image deblurring framework with stepwise feature refinement module

Article Computer Science, Artificial Intelligence

Vision Transformers for Single Image Dehazing

Yuda Song et al.

Summary: Image dehazing is a low-level vision task to estimate haze-free images. Convolutional neural network methods dominate this task, but vision Transformers haven't made a breakthrough. This study introduces DehazeFormer, an improved version of Swin Transformer, with modified normalization layer, activation function, and spatial information aggregation. Multiple variants of DehazeFormer were trained and shown to be effective.

IEEE TRANSACTIONS ON IMAGE PROCESSING (2023)

Add to Collection

Article Computer Science, Information Systems

DeepRecog: Threefold underwater image deblurring and object recognition framework for AUV vision systems

M. V. Pranav et al.

Summary: This paper proposes an integrated underwater image deblurring and object recognition framework DeepRecog for AUV vision systems, providing efficient real-time underwater detection and recognition. The framework utilizes CNNs, adaptive filters, and transformative filters to achieve image deblurring, and achieves high precision through an ensemble object detection and recognition module.

MULTIMEDIA SYSTEMS (2022)

Add to Collection

Article Computer Science, Artificial Intelligence

Surface-Aware Blind Image Deblurring

Jun Liu et al.

Summary: Blind image deblurring is a difficult problem that requires proper prior knowledge. The surface-aware strategy proposed in this paper facilitates blur kernel estimation and outperforms other methods in experiments.

IEEE TRANSACTIONS ON PATTERN ANALYSIS AND MACHINE INTELLIGENCE (2021)

Add to Collection

Article Computer Science, Information Systems

Attention-adaptive and deformable convolutional modules for dynamic scene deblurring

Lei Chen et al.

Summary: This study investigates two aspects of network architecture design for dynamic scene deblurring: learning blur characteristics and location, and handling geometric variations. By designing an attention-adaptive module and a deformable convolutional module, the performance can be significantly improved to adapt to the changes in dynamic scenes. Incorporating these proposed modules into existing models shows promising results compared to state-of-the-art methods.

INFORMATION SCIENCES (2021)

Add to Collection

Article Computer Science, Artificial Intelligence

Blind Motion Deblurring Super-Resolution: When Dynamic Spatio-Temporal Learning Meets Static Image Understanding

Wenjia Niu et al.

Summary: Single-image super-resolution and multi-frame super-resolution are two ways to improve the quality of low-resolution images, with the former handling each image independently and the latter modeling temporal dependency. However, multi-frame methods rely on neighboring frames that are not always available in the real world, and camera shake can easily cause motion blur. To address these challenges, a Blind Motion Deblurring Super-Resolution Network (BMDSRNet) is proposed to learn dynamic spatio-temporal information from single static motion-blurred images and outperform recent state-of-the-art methods.

IEEE TRANSACTIONS ON IMAGE PROCESSING (2021)

Add to Collection

Article Computer Science, Artificial Intelligence

Structure-Aware Motion Deblurring Using Multi-Adversarial Optimized CycleGAN

Yang Wen et al.

Summary: This paper presents an unsupervised image deblurring method based on a multi-adversarial optimized cycle-consistent generative adversarial network (CycleGAN), which maintains structural information better through multi-adversarial learning and structure-aware mechanism.

IEEE TRANSACTIONS ON IMAGE PROCESSING (2021)

Add to Collection

Article Engineering, Electrical & Electronic

Polyblur: Removing Mild Blur by Polynomial Reblurring

Mauricio Delbracio et al.

Summary: The proposed method is a highly efficient blind restoration method aimed at removing mild blur in natural images. It estimates image blur and compensates for it by combining multiple applications of the estimated blur in a principled way, outperforming traditional and modern blind deblurring methods in the context of mild blur.

IEEE TRANSACTIONS ON COMPUTATIONAL IMAGING (2021)

Add to Collection

Article Engineering, Electrical & Electronic