Temporal Image Denoising Using Multi-Stage Frame Fusion

Resolve Bottlenecks,
Find Innovative SolutionsGenerate Solutions

Solution Overview

Problem

Image quality is degraded due to faulty or non-functional pixels in high-density image sensors, necessitating effective noise removal methods that preserve image details.

Innovation Solution

A three-stage gradual temporal fusion deep learning network using recurrent neural networks to remove noise from images, incorporating information from previous frames to improve denoising accuracy and preserve image details.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Manufacturing precision

If high-density image sensors are used to achieve smaller pixel sizes and greater integration density, then image resolution and integration capacity are improved, but faulty or non-functional pixels occur leading to degraded image quality

Engineering Contradiction:

Improvepixel size and integration densityVSAvoidimage quality

Core Design Contradiction:

Manufacturing precisionVSReliability

Solution Approach 1:

The patent introduces a denoising module as an intermediary component that processes pixel data from the high-density image sensor. This module uses machine learning algorithms to identify and correct faulty pixel data, acting as a mediator between the sensor and the final image output, thereby maintaining image quality despite the presence of defective pixels in high-density configurations

Inventive Principle:

Principle #24Intermediary (Mediator)

Solution Approach 2:

The patent replaces traditional mechanical or algorithmic noise reduction methods with a machine learning-based denoising module. This neural network-based system learns from training data to distinguish between actual image features and noise artifacts, providing more effective noise removal while preserving image details compared to conventional approaches

Inventive Principle:

Principle #28Mechanics substitution (Replace mechanical system)

2Reliability

If noise removal is performed to correct faulty pixel data, then image quality is improved, but image details may be lost

Engineering Contradiction:

Improveimage qualityVSAvoidimage details

Core Design Contradiction:

ReliabilityVSLoss of information

Solution Approach 1:

The denoising module applies local quality processing by analyzing small patches of pixel data independently. The machine learning model evaluates each local region to determine the appropriate amount of denoising to apply, preserving edges and fine details in high-contrast areas while aggressively removing noise in uniform regions, thereby maintaining overall image quality without losing important details

Inventive Principle:

Principle #3Local quality

Solution Approach 2:

The system implements feedback mechanisms where the denoised output is compared with the original input, and the difference is used to refine the denoising process. The machine learning model adjusts its processing based on feedback from multiple iterations, ensuring that noise removal does not inadvertently remove important image details while continuously improving image quality

Inventive Principle:

Principle #23Feedback

Data Source

PatentUS12475535B2Image processing device, electronic device having the same, and operating method thereof

Publication Date: 2025.11.18 SAMSUNG ELECTRONICS CO LTD

AI summary

A method includes acquiring a current frame, a noise dispersion map for the current frame, and a first denoised frame for a previous frame; generating a weighted first denoised frame based on the noise dispersion map, the current frame, and the first denoised frame using a first neural network; generating an initial fused image based on the current frame and the weighted first denoised frame using a second neural network; and generating a second denoised frame for the current frame based on the initial fused image using a third neural network.