Neural Network Low-Light Image Enhancement

Resolve Bottlenecks,
Find Innovative SolutionsGenerate Solutions

Solution Overview

Problem

Conventional imaging techniques produce sub-optimal images in low-light conditions due to low photon counts and high noise, with existing methods like long exposure times causing blur, large lenses being impractical, exposure bracketing yielding unsatisfactory results, burst imaging leading to misalignment artifacts, and flash photography introducing unwanted reflections and shadows.

Innovation Solution

A neural network architecture trained with a multi-criterion loss function that learns to transform raw image sensor data into well-exposed images by utilizing both pixel-level and feature-level information, eliminating noise and artifacts, and incorporating a contrast correction module to enhance image quality.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Illumination intensity

If long exposure times are used to capture more light in low-light conditions, then image brightness is improved, but image sharpness deteriorates due to camera shake and object movement

Engineering Contradiction:

Improveimage brightnessVSAvoidimage sharpness

Core Design Contradiction:

Illumination intensityVSManufacturing precision

Solution Approach 1:

The system performs preliminary actions by capturing multiple images at different exposure levels before final processing. Multiple images are captured in quick succession with varying exposure times, and then processed using exposure bracketing techniques to combine the best-exposed regions from each image, achieving both brightness and sharpness without requiring a single long exposure

Inventive Principle:

Principle #10Preliminary action

Solution Approach 2:

The system merges multiple images captured at different exposure levels into a single final image. By combining the well-exposed regions from multiple short-exposure images, the system achieves the brightness enhancement of long exposure while maintaining the sharpness of short exposure, effectively resolving the contradiction between these two parameters

Inventive Principle:

Principle #5Merging (Combining)

2Illumination intensity

If a large lens with fast aperture speed is used to capture more light, then image brightness is improved, but device complexity and size increase

Engineering Contradiction:

Improveimage brightnessVSAvoidlens size

Core Design Contradiction:

Illumination intensityVSDevice complexity

Solution Approach 1:

The system replaces the mechanical solution of using a large lens with fast aperture (physical/optical system) with a computational approach using multiple image captures and processing algorithms. Instead of relying on optical hardware to gather more light, the system uses software-based exposure bracketing and image fusion techniques to achieve brightness enhancement, thereby avoiding the need for large, complex lens systems

Inventive Principle:

Principle #28Mechanics substitution (Replace mechanical system)

Solution Approach 2:

The system changes the exposure time parameter across multiple captures rather than changing the aperture parameter. By varying exposure duration in software control and combining results computationally, the system achieves the light-gathering effect of a fast aperture without physically requiring a large lens, thus reducing device complexity and size

Inventive Principle:

Principle #35Parameter changes

3Adaptability or versatility

If exposure bracketing is used to capture images with varying shutter speeds, then adaptability to different lighting conditions is improved, but image quality deteriorates in low-light conditions

Engineering Contradiction:

Improvelighting condition adaptabilityVSAvoidimage quality

Core Design Contradiction:

Adaptability or versatilityVSManufacturing precision

Solution Approach 1:

The system incorporates feedback mechanisms by evaluating the quality of captured images and adjusting subsequent capture parameters. The processing algorithm analyzes the captured images to identify well-exposed regions and uses this information to guide the fusion process, ensuring that only high-quality regions are combined, thereby maintaining image quality while preserving adaptability to different lighting conditions

Inventive Principle:

Principle #23Feedback

4Illumination intensity

If burst imaging and fusion techniques are used to capture multiple images, then image brightness is improved, but alignment accuracy deteriorates causing artifacts

Engineering Contradiction:

Improveimage brightnessVSAvoidimage alignment accuracy

Core Design Contradiction:

Illumination intensityVSMeasurement precision

Solution Approach 1:

The system performs preliminary alignment and registration of multiple captured images before fusing them. By pre-aligning the images using feature matching and transformation algorithms, the system ensures accurate correspondence between images from different exposures, preventing misalignment artifacts while still achieving brightness enhancement through the fusion of multiple captures

Inventive Principle:

Principle #10Preliminary action

Data Source

PatentUS11037278B2Systems and methods for transforming raw sensor data captured in low-light conditions to well-exposed images using neural network architectures

Publication Date: 2021.06.15 INCEPTION AI IP LTD

AI summary

This disclosure relates to improved techniques for generating images from raw image sensor data captured in low-light conditions without the use of flash photography. The techniques described herein utilize a neural network architecture to transform the raw image sensor data into well-exposed images. The neural network architecture can be trained using a multi-criterion loss function that jointly models both pixel-level and feature-level properties of the images. The images output by the neural network architecture can be provided to a contrast correction module that enhances the contrast of the images.