ExposureNet: Mobile camera exposure parameters autonomous control for blur effect prevention

Abstract

AbstractThe quality of images we perceive visually is heavily impacted by the settings used for camera exposure. When these settings are imbalanced, it can result in an undesired prominent phenomenon known as blur effects. To address this problem, an ExposureNet project has been undertaken, which aims to develop an autonomous camera exposure settings control system for blur effects prevention. The proposed ExposureNet model is a CNN/Transformer hybrid neural structure, created and trained in a comprehensive manner to effectively predict the ideal exposure settings based on the semantic features of the scene being captured. This system is designed to learn the necessary steps for processing, such as identifying relevant scene features, using only two camera exposure parameters (shutter speed (SHS) and ISO) as training signals. As a result, this system can associate the semantic features of a scene with the appropriate exposure parameter adjustments, customized to the scene's dynamics and lighting conditions. By simultaneously optimizing all processing steps and bypassing traditional post‐processing stages, the proposed system is designed to achieve faster performance, reduced computational cost, and lower power consumption. Experimental results demonstrate that the proposed system significantly outperforms existing methods and achieves cutting‐edge performance.