Neural Network Long-Exposure Image Synthesis
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Mobile devices struggle to capture high-quality long-exposure photographs due to lack of necessary hardware and limited settings, resulting in blurry and unsatisfactory images, and conventional systems that attempt to create long-exposure photos from videos often produce unrealistic and low-resolution results.
Innovation Solution
An image exposure transformation network is developed, utilizing a generator neural network and adversarial discriminator to transform short-exposure images into realistic long-exposure images, incorporating optical flow prediction and appearance guided attention networks to improve image quality and accuracy.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If conventional systems create long-exposure photographs from video sequences, then long-exposure effect is achieved, but image resolution and realism deteriorate due to artifacts and pixel movements
Solution Approach 1:
The patent replaces the mechanical/video-based long-exposure system with a neural network-based image processing system. The GAN transforms a single short-exposure image into a long-exposure image through learned transformations, eliminating the need for video sequences and manual frame-by-frame processing while maintaining high resolution and realism.
Solution Approach 2:
The patent changes the fundamental parameter from video frame rate and exposure time to neural network transformation parameters. The GAN learns to map short-exposure images to long-exposure images by adjusting network weights and parameters during training, enabling continuous optimization of output quality without being constrained by video capture parameters.
2Manufacturing precision
If mobile devices use additional hardware like tripods and remote triggers to capture long-exposure photographs, then image quality improves, but device complexity and accessibility worsen
Solution Approach 1:
The patent enables the mobile device to perform long-exposure photography functions using its own existing camera and processor. The neural network runs locally on the device, transforming short-exposure images captured by the device's native camera into long-exposure images without requiring external tripods, remote triggers, or specialized hardware attachments.
Solution Approach 2:
The patent makes the mobile device camera universally capable of both short-exposure and long-exposure photography using the same hardware. The neural network-based solution allows the device to switch between exposure types through software processing alone, eliminating the need for dedicated long-exposure hardware components.
3Reliability
If conventional systems process video sequences to create long-exposure photographs, then exposure transformation is achieved, but processing time and computational resources increase
Solution Approach 1:
The patent extracts the essential long-exposure effect from the complex video processing pipeline and encapsulates it in a dedicated neural network model. Instead of processing entire video sequences frame-by-frame, the GAN extracts and applies only the necessary transformation to a single image, dramatically reducing computational overhead and processing time.
Solution Approach 2:
The patent performs preliminary training of the neural network offline using video data and ground truth long-exposure images. Once trained, the network contains pre-learned transformation patterns that can be rapidly applied to new images without requiring real-time video processing, enabling fast inference while maintaining transformation accuracy.
Data Source
AI summary
The present disclosure relates to training and utilizing an image exposure transformation network to generate a long-exposure image from a single short-exposure image (e.g., still image). In various embodiments, the image exposure transformation network is trained using adversarial learning, long-exposure ground truth images, and a multi-term loss function. In some embodiments, the image exposure transformation network includes an optical flow prediction network and/or an appearance guided attention network. Trained embodiments of the image exposure transformation network generate realistic long-exposure images from single short-exposure images without additional information.