Burst Image Restoration and Enhancement

Area of Science:

• Computer Vision
• Image Processing

Background:

• Burst Image Restoration (BIR) aims to improve image quality by merging information from multiple frames.
• Inter-frame misalignments in burst images cause artifacts like ghosting and zippers, hindering effective restoration.

Purpose of the Study:

• To develop a novel approach, BIPNet, for effective burst image processing.
• To address challenges posed by inter-frame misalignments and preserve scene details.

Main Methods:

• BIPNet generates pseudo-burst features by combining information from all burst frames.
• Edge-boosting feature alignment corrects inter-frame misalignments relative to a reference frame.
• Adaptive-progressive feature up-sampling enhances details through multi-stage upscaling and fusion.

Main Results:

• BIPNet effectively mitigates ghosting and zipper artifacts.
• The method significantly outperforms existing techniques in burst super-resolution, low-light enhancement, and denoising.
• Achieved superior performance across multiple image restoration tasks.

Conclusions:

• BIPNet offers a robust solution for burst image restoration by addressing misalignment issues.
• The proposed pseudo-burst mechanism and adaptive up-sampling enhance image quality and detail preservation.