Low-Light Frame Enhancement Using Key and Dependent Neural Networks

Resolve Bottlenecks,
Find Innovative SolutionsGenerate Solutions

Solution Overview

Problem

Existing image processing systems face challenges in performing low-light image enhancement without additional hardware, leading to accuracy losses and increased complexity, cost, and power consumption during low-light scenarios.

Innovation Solution

A machine learning system utilizing a first neural network for enhancing low-light key frames and a second neural network for enhancing dependent frames, leveraging a hidden state from the first network to improve the processing efficiency and accuracy of low-light images.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Device complexity

If existing image processing systems perform low-light image enhancement without additional hardware, then the system complexity and cost are reduced, but accuracy losses occur

Engineering Contradiction:

Improvesystem complexityVSAvoidenhancement accuracy

Core Design Contradiction:

Device complexityVSMeasurement precision

Solution Approach 1:

The image enhancement task is segmented into two distinct phases: key frame enhancement using a first neural network, and dependent frame enhancement using a second neural network. This segmentation allows the system to apply different processing strategies to different frames, improving overall accuracy while managing computational complexity efficiently.

Inventive Principle:

Principle #1Segmentation

Solution Approach 2:

The system performs preliminary enhancement on key frames first, generating enhanced representations that serve as prior information for subsequent dependent frame enhancement. This preliminary action on select frames prepares the system to process remaining frames more efficiently with maintained accuracy.

Inventive Principle:

Principle #10Preliminary action

2Ease of manufacture

If existing image processing systems perform low-light image enhancement without additional hardware, then the hardware cost is reduced, but accuracy losses occur

Engineering Contradiction:

Improvehardware costVSAvoidenhancement accuracy

Core Design Contradiction:

Ease of manufactureVSMeasurement precision

Solution Approach 1:

The system creates computational copies of enhancement capabilities through software-based neural networks instead of requiring additional physical hardware. The first and second neural networks serve as virtual copies of enhancement functionality, eliminating the need for specialized hardware while maintaining enhancement accuracy.

Inventive Principle:

Principle #26Copying

3Measurement precision

If existing image processing systems perform low-light image enhancement, then image quality is improved, but power consumption increases

Engineering Contradiction:

Improveimage qualityVSAvoidpower consumption

Core Design Contradiction:

Measurement precisionVSUse of energy by moving object

Solution Approach 1:

The power-intensive enhancement task is segmented across two neural networks processed in sequence. The first network processes key frames, and the second network processes dependent frames using information from the first, distributing computational load and reducing peak power consumption while maintaining image quality.

Inventive Principle:

Principle #1Segmentation

Solution Approach 2:

The system discards redundant processing for dependent frames by leveraging the hidden state from key frame enhancement. Instead of fully reprocessing each dependent frame, the system recovers useful information from the key frame representation, reducing computational power requirements while preserving enhancement quality.

Inventive Principle:

Principle #34Discarding and recovering

4Measurement precision

If existing image processing systems perform low-light image enhancement, then image quality is improved, but processing time increases

Engineering Contradiction:

Improveimage qualityVSAvoidprocessing time

Core Design Contradiction:

Measurement precisionVSLoss of time

Solution Approach 1:

The enhancement process is segmented into key frame processing and dependent frame processing. This segmentation allows the system to apply full enhancement only to select key frames while using accelerated processing for dependent frames, reducing overall processing time while maintaining quality.

Inventive Principle:

Principle #1Segmentation

Solution Approach 2:

The system discards redundant computation for dependent frames by recovering enhancement information from key frame hidden states. This approach eliminates unnecessary reprocessing steps, significantly reducing processing time while preserving image enhancement quality through information reuse.

Inventive Principle:

Principle #34Discarding and recovering

Data Source

PatentUS12488425B2Low-light image enhancement using key frame and dependent frame neural networks

Publication Date: 2025.12.02 QUALCOMM INC

AI summary

Systems and techniques are provided for processing image data. For instance, a process can include classifying a first image as a key frame, based on a difference between the first image and a previous image, wherein the first image and the previous image are included in a plurality of images. A first machine learning network can generate an enhanced key frame image corresponding to the first image and a hidden state output associated with the enhanced key frame image. A second image of the plurality of images can be classified as a dependent frame, based on a similarity between the second image and the first image. A second machine learning network can generate an enhanced dependent frame image corresponding to the second image, wherein the enhanced dependent frame image is based on the second image and the hidden state output associated with the enhanced key frame image.