Neuromorphic Imaging With Joint Image Deblurring and Event Denoising.

Area of Science:

• Computer Vision
• Neuromorphic Engineering
• Image Processing

Background:

• Neuromorphic imaging offers high temporal precision but suffers from noisy events and blurry images of fast-moving objects.
• These data deficiencies hinder effective human observation and machine processing.
• Events capture object edges and motion details, complementing latent sharp images.

Purpose of the Study:

• To develop a unifying algorithm for joint reconstruction of blur-free images and noise-robust events from neuromorphic imaging data.
• To leverage the complementary nature of event streams and intensity images for improved data quality.
• To enhance the robustness and efficiency of neuromorphic data processing.

Main Methods:

• An iterative coarse-to-fine algorithm was developed to fuse event and image data.
• Event-regularized priors were used for blind deblurring, incorporating high-frequency structures and dynamic features.
• Image gradients provided supervision for neuromorphic noise removal.

Main Results:

• The synergistic approach significantly improved reconstruction quality for both motion-blurred images and noisy event streams.
• The method demonstrated superior robustness across varying illumination, contrast, and motion magnitudes.
• The algorithm requires fewer events and offers competitive computational efficiency.

Conclusions:

• The proposed method effectively addresses the limitations of raw neuromorphic data, producing high-quality images and clean event streams.
• This technique enhances the utility of neuromorphic sensing for tasks requiring accurate motion analysis and reasoning.
• The solution is suitable for resource-limited computing environments, promoting wider adoption of neuromorphic imaging.