EvTexture++: Event-Driven Texture Enhancement for Video Super-Resolution.

Area of Science:

• Computer Vision
• Image Processing
• Deep Learning

Background:

• Event-based cameras offer high temporal resolution and dynamic range.
• Existing video super-resolution (VSR) methods utilize events for motion estimation.
• A gap exists in leveraging event data for texture enhancement in VSR.

Purpose of the Study:

• To propose EvTexture++, the first event-driven framework for texture enhancement in VSR.
• To improve texture recovery and temporal consistency in VSR using event data.
• To offer a plug-and-play module for enhancing existing VSR models.

Main Methods:

• EvTexture++ utilizes high-frequency spatiotemporal details from events for texture recovery.
• An iterative texture enhancement module progressively refines textures using event information.
• A temporal texture alignment module uses event-guided flow to ensure inter-frame consistency.

Main Results:

• EvTexture++ achieves state-of-the-art performance on multiple datasets.
• Integration with existing VSR models yields significant improvements, up to 1.55 dB PSNR.
• The framework effectively enhances texture details and reduces flickering in VSR outputs.

Conclusions:

• EvTexture++ demonstrates the efficacy of event data for texture enhancement in VSR.
• The proposed framework improves both intra-frame texture recovery and inter-frame temporal consistency.
• EvTexture++ offers a versatile solution for advancing VSR technology.