Non-uniform Exposure Prediction Pipeline for Burst Image Restoration
Yongkee Lee, Woohyeok Kim, Sunghyun Cho · Journal of the Korea Computer Graphics Society · 2025
Burst image restoration is a technique that synthesizes multiple images captured consecutively over a short time period to restore a single high-quality image. While numerous studies have been conducted in this field, most rely on uniform exposure settings or predefined non-uniform configurations (e.g., fixed-ratio exposure brackets). In such scenarios, burst images often exhibit similar levels of blur and noise, leading to insufficient complementary information and a failure to adequately capture the characteristics of the imaging environment. To address these limitations, we propose a non-uniform exposure prediction pipeline that dynamically optimizes exposure parameters based on scene conditions and restoration requirements. The proposed pipeline facilitates environment-aware burst image capture, specifically tailored for restoration tasks, thereby significantly enhancing image quality in low-light conditions.