Image Restoration Quantization with Gamma Correction for Low-Light Quality
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing image restoration deep learning models suffer from significant image quality degradation during quantization, particularly in low-light environments, due to inadequate preprocessing based on pixel distribution, leading to uneven bit allocation and increased noise in dark areas.
Innovation Solution
An image restoration method that applies gamma correction to adjust pixel value distribution, followed by quantization and inverse conversion to minimize quality loss, using a conversion parameter determined by illuminance and task target, and employing a quantized image restoration model.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Device complexity
If quantization is applied to deep learning model for image restoration, then model complexity and computational cost are reduced, but image quality degrades significantly especially in low-light environments
Solution Approach 1:
The patent applies gamma correction as a preliminary action before quantization to adjust the pixel value distribution. By performing this preprocessing step, the patent prepares the input data in a way that makes it more suitable for quantization, thereby reducing the negative impact on image quality while maintaining the benefits of quantization for model complexity reduction.
Solution Approach 2:
The patent changes the parameter distribution of the input image by applying gamma correction with specific gamma values (e.g., 0.45 for low-light conditions). This parameter transformation redistributes the pixel values to optimize the quantization process, ensuring that critical image information is preserved even after quantization reduces model complexity.
2Speed
If standard quantization is used without preprocessing, then processing speed increases, but noise increases in dark areas due to uneven bit allocation
Solution Approach 1:
The patent applies gamma correction to change the parameter distribution of pixel values before quantization. This transformation ensures that dark areas with low pixel values are properly scaled, leading to more uniform bit allocation during quantization and reducing noise in dark regions while maintaining processing speed.
Solution Approach 2:
By performing gamma correction as a preliminary action before quantization, the patent prepares the data to prevent noise formation in dark areas. This preprocessing step ensures that when quantization is applied for fast processing, the bit allocation is optimized to avoid amplifying noise in low-brightness regions.
3Manufacturing precision
If gamma correction is applied to adjust pixel distribution, then bit allocation becomes more uniform and image quality is preserved, but additional processing steps are required
Solution Approach 1:
The patent applies gamma correction to change the parameter distribution of pixel values, which uniformly distributes bit allocation across different brightness levels. This parameter transformation preserves image quality by ensuring that both dark and bright areas maintain appropriate detail after quantization, despite adding a processing step.
Data Source
AI summary
A processor-implemented method with image restoration includes obtaining an input image, generating a correction image by converting the input image based on a conversion parameter that converts a pixel value distribution of the input image, generating an output image by inputting the correction image to a quantized image restoration model, and generating an enhanced image by inversely converting the output image based on an inverse conversion parameter corresponding to the conversion parameter.