Image Pyramid Denoising for Low-Light Noise and Edge Preservation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing image processing systems struggle to effectively remove noise, particularly in low-light conditions, due to hardware limitations and the difficulty in using large filters, leading to challenges in distinguishing pixel signals from noise and edge information.
Innovation Solution
A system-on-chip and electronic device employing an image pyramid structure that includes a pyramid generation module, denoising process module, and pyramid reconstruction module to generate and process Gaussian and Laplacian pyramids, enabling accurate edge identification and noise removal through multi-scaling denoising methods.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If a large filter is used to process low-frequency noise in spatial domain, then noise reduction performance is improved, but device complexity and hardware implementation difficulty increase
Solution Approach 1:
The patent segments the image processing task into multiple frequency bands using wavelet transform, dividing the original image into approximation coefficients and detail coefficients at different scales. This segmentation allows independent processing of low-frequency noise without requiring a large spatial filter, thereby reducing hardware complexity while maintaining noise reduction performance.
Solution Approach 2:
The patent transitions from spatial domain filtering to frequency domain processing by applying wavelet transform. This dimensional change from spatial to frequency domain enables effective low-frequency noise removal through coefficient thresholding rather than large spatial filters, resolving the contradiction between noise reduction performance and hardware complexity.
2Reliability
If noise processing is performed to improve image quality, then image clarity is improved, but processing time and power consumption increase
Solution Approach 1:
The patent extracts noise components from the image by separating them into specific wavelet coefficient bands (detail coefficients). By isolating and processing only the noisy frequency components rather than the entire image, the processing time is significantly reduced while maintaining image quality improvement.
Solution Approach 2:
The patent changes the processing parameters by applying different thresholding strategies to different wavelet coefficient bands. High-frequency detail coefficients undergo aggressive thresholding to remove noise, while low-frequency approximation coefficients are preserved. This parameter differentiation enables efficient noise processing that reduces overall processing time while maintaining image quality.
3Reliability
If aggressive noise removal is applied, then noise reduction performance is improved, but edge information and pixel signals are distorted or lost
Solution Approach 1:
The patent applies local quality by treating different frequency bands with different processing intensities. Detail coefficients (high-frequency components containing edges) receive adaptive thresholding that preserves significant variations, while approximation coefficients (low-frequency components containing smooth regions) receive more aggressive noise suppression. This local differentiation in processing quality maintains edge information while removing noise.
Solution Approach 2:
The patent incorporates feedback mechanisms in the wavelet denoising process by using the reconstructed image from previous iterations to guide subsequent processing. The feedback loop allows the system to preserve edge information detected in earlier stages while continuing to remove noise, preventing information loss while maintaining noise reduction performance.
Data Source
AI summary
A system-on-chip, an electronic device, and an operating method of a processor for reducing noise in an image based on an image pyramid are provided. The system-on-chip includes a pyramid generation module outputting a Gaussian image of a highest layer and a Laplacian pyramid, based on an input image, a denoising process module denoising the Gaussian image of the highest layer and outputting a denoised image of the highest layer, and a pyramid reconstruction module receiving the Laplacian pyramid and the denoised image of the highest layer, generating a denoised image and edge grade information for each layer based on a Laplacian image of each layer, a denoised image of an upper layer that is higher than each layer, and edge grade information of the upper layer, and outputting an output image.