Bayer-Domain Wavelet Denoising for Low-Light Image Enhancement
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Low-light images suffer from noise, loss of detail, color distortion, reduced dynamic range, and digital artifacts such as color noise, banding, and posterization, making them challenging to enhance effectively.
Innovation Solution
Perform denoising preprocessing on low-light images using wavelet decomposition and neural networks in the Bayer domain before inputting the image into the ISP pipeline, creating frequency domain subimages that are processed by corresponding neural networks to improve image quality.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Illumination intensity
If the aperture is enlarged to allow additional light to reach the image sensor, then the amount of light reaching the sensor is improved, but the depth of field is reduced, causing at least part of the image to appear out of focus
Solution Approach 1:
The patent applies denoising preprocessing to the raw low-light image before it enters the ISP pipeline. By performing noise reduction in advance on the Bayer domain data, the system improves the signal-to-noise ratio without requiring optical adjustments that would compromise depth of field. This preliminary action allows the aperture to remain open for light gathering while avoiding the focus issues that would result.
Solution Approach 2:
The patent replaces mechanical/optical solutions (aperture adjustment) with a digital signal processing approach. Instead of changing the physical aperture size to improve light capture, the system uses wavelet decomposition and neural network-based denoising algorithms to process the image data digitally, achieving improved image quality without mechanical adjustments that would affect depth of field.
2Illumination intensity
If the exposure time is increased to enable additional light to reach the sensor, then the amount of light reaching the sensor is improved, but the probability of undesired motion blur is increased
Solution Approach 1:
The patent performs denoising preprocessing on the raw image data before ISP processing. By addressing noise issues in advance through wavelet decomposition and neural network filtering, the system can achieve acceptable image quality with shorter exposure times, thereby reducing motion blur while still capturing sufficient light.
Solution Approach 2:
The patent replaces the mechanical/optical approach of increasing exposure time with a digital signal processing solution. Instead of extending exposure duration to gather more light, the system uses advanced denoising algorithms to process the captured signal, achieving improved image quality without the motion blur that would result from longer exposures.
3Illumination intensity
If the ISO is increased to improve sensitivity to light, then the sensor's sensitivity to light is improved, but the amount of digital noise in the image is increased
Solution Approach 1:
The patent extracts and removes noise from the low-light image through wavelet decomposition and neural network-based denoising. By separating the noise component from the actual image signal in the frequency domain, the system can eliminate digital noise while preserving the useful image information, effectively solving the ISO noise problem through selective extraction and removal.
Solution Approach 2:
The patent replaces the mechanical/electrical approach of increasing ISO sensitivity with a digital signal processing solution. Instead of amplifying the sensor signal (which amplifies noise), the system uses wavelet decomposition and neural networks to process the raw data, achieving improved light sensitivity效果 without the accompanying digital noise through intelligent signal separation and filtering.
4Measurement precision
If wavelet decomposition and neural network processing are applied to denoise the image, then the noise reduction and detail extraction are improved, but the processing complexity is increased
Solution Approach 1:
The patent segments the image processing task into distinct stages: wavelet decomposition into frequency subbands, followed by neural network processing of specific subbands, and finally inverse wavelet transformation. This segmentation allows complex denoising to be broken into manageable steps, each handling specific frequency components, thereby improving noise reduction effectiveness while making the overall complex process more controllable and efficient.
Data Source
AI summary
A system and method are disclosed for low-light image enhancement using denoising preprocessing with wavelet decomposition AI-based techniques to enhance image quality of low-light images. Subsampled images are created from a raw input image. A wavelet decomposition process is performed on each subimage to create multiple frequency domain subimages. Each frequency domain subimage is input into a corresponding neural network. The output of each corresponding network is input to an inverse wavelet module. The output of the inverse wavelet module is a denoised image that is input to an image signal processing pipeline, where additional processing may be performed on the denoised image.