Neural Network Denoising for Low-Light Camera Image Quality
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Smart terminal devices struggle to capture clear images in low-illumination environments due to poor photosensitivity of camera sensors, resulting in noisy and blurred images.
Innovation Solution
An image display method that processes N frames of raw images using a neural network model to improve denoising effects, where the neural network model uses an image with noise lower than a target threshold as an output objective, and the images are displayed on a viewfinder frame, without prior processing by an ISP or accelerator, preserving the original noise form.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If camera sensor size and aperture are increased to improve photosensitivity in low-illumination scenarios, then image quality improves, but device size and complexity increase
Solution Approach 1:
The patent changes the parameter of image processing by applying a neural network model that specifically targets noise reduction in low-illumination scenarios. Instead of hardware modifications, the solution transforms the software processing parameters to achieve better image quality from the same hardware input.
Solution Approach 2:
The patent replaces the mechanical approach of increasing sensor size and aperture with an algorithmic approach using neural network-based denoising. This substitution achieves the same goal of improving low-light image quality without the physical constraints of hardware scaling.
2Reliability
If N frames of raw images are processed using a neural network model to reduce noise, then image denoising effect improves, but processing time and computational complexity increase
Solution Approach 1:
The patent applies preliminary denoising processing to N frames of raw images before final image synthesis. By performing noise reduction early in the processing pipeline on the raw data, the neural network model prevents noise accumulation and enables more effective subsequent processing with fewer computational iterations.
Solution Approach 2:
The patent segments the image processing task into distinct stages: first capturing N frames of raw images, then applying neural network denoising to each frame or combined data, and finally synthesizing the result. This segmentation allows optimized processing at each stage rather than treating the entire process as a single computational burden.
3Reliability
If multiple frames of raw images are captured and processed to reduce noise, then image quality improves, but frame rate and responsiveness decrease
Solution Approach 1:
The patent processes only N specific frames (where N is a small integer greater than or equal to 2) rather than continuously processing all incoming frames. This partial processing approach applies denoising selectively to achieve quality improvement while maintaining acceptable frame rates for real-time preview and capture.
Data Source
AI summary
An image display method and device are provided. The method is applied to an electronic device having a display screen and a camera. The method includes: detecting a first operation of turning on the camera by a user; displaying a photographing interface on the display screen in response to the first operation, where the photographing interface includes a viewfinder frame including a first image; detecting a second operation of the camera indicated by the user; and displaying a second image in the viewfinder frame in response to the second operation, where the second image is an image obtained by processing N frames of raw images captured by the camera; a neural network model is applied to a processing process, and the neural network model uses an image whose noise is lower than a target threshold as an output objective; and N is an integer greater than or equal to 2.