Saliency-Based Image Denoising for Detail Preservation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Traditional image denoising techniques face challenges in efficiently removing noise from images, particularly when noise characteristics are unknown or time-varying, leading to either under- or over-denoising, which affects image quality and bitrate savings, and often uniformly remove noise across the image, compromising salient regions.
Innovation Solution
The method employs a saliency map to determine noise levels for different regions of an image based on their perceptual importance, using saliency scores to guide denoising and adjust encoding parameters, thereby preserving details in salient regions while over-denoising less important areas.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Object-affected harmful factors
If traditional uniform denoising is applied across the entire image, then noise removal is achieved, but salient regions lose important details and textures
Solution Approach 1:
The patent applies different denoising strengths to different regions of the image based on their saliency scores. Salient regions (with high saliency scores) receive mild denoising to preserve details and textures, while non-salient regions (with low saliency scores) receive strong denoising to remove noise effectively. This is achieved by computing a saliency map and using it to modulate the denoising filter strength across the image.
Solution Approach 2:
The image is effectively segmented into salient and non-salient regions through the saliency map computation. This segmentation allows the denoising process to be applied selectively - strong denoising for background/non-salient areas and weak denoising for foreground/salient areas, thereby resolving the contradiction between noise removal and detail preservation.
2Object-affected harmful factors
If strong denoising is applied to remove all noise, then noise levels decrease, but image quality and perceptual details are compromised
Solution Approach 1:
Instead of applying uniform strong denoising, the patent uses local quality adjustment by modulating denoising strength according to saliency. Non-salient regions receive strong denoising for low noise levels, while salient regions receive weak denoising to maintain image quality and perceptual details. The saliency map guides this localized adjustment.
3Manufacturing precision
If weak denoising is applied to preserve all details, then detail preservation is achieved, but noise remains in non-salient regions increasing bitrate
Solution Approach 1:
The patent applies local quality adjustment by using saliency-based modulation of denoising strength. Salient regions receive weak denoising to preserve details, while non-salient regions receive strong denoising to remove noise. This selective approach reduces the overall bitrate by removing noise from non-salient areas without compromising important details in salient regions.
4Productivity
If selective denoising based on saliency is implemented, then compression performance improves, but computational complexity increases
Solution Approach 1:
The patent uses segmentation through saliency map computation to divide the image into salient and non-salient regions. This segmentation enables selective denoising that improves compression performance by removing noise from non-salient regions. The computational complexity is managed by using efficient saliency detection algorithms and applying denoising only where necessary.
Solution Approach 2:
The patent changes the denoising parameter (filter strength) based on the saliency parameter. By modulating the denoising strength according to saliency scores, the system achieves improved compression performance through selective noise removal, while the parameter changes are computed efficiently using the pre-calculated saliency map.
Data Source
AI summary
Image denoising includes obtaining a saliency map for an image. The saliency map includes respective saliency scores for pixels of the image. Respective noise levels are assigned to the pixels using the respective saliency scores to obtain a noise level map. The image is denoised using the noise level map to obtain a denoised image. The denoised image is output, such as to a display or a storage device.