Near-True-View Medical Video Processing With Single-Pass SDDTT
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing video processors face challenges in producing high-quality images from videoscopes due to noise and illumination issues, which are exacerbated by factors like moisture, occlusion, and conflicting effects of denoising and sharpening, leading to increased costs and limitations in improving image quality, especially for single-use devices.
Innovation Solution
A single denoising and feature/edge detection trained network (SDDTT) processes images in real-time to generate near real-view images, enhancing image quality while reducing processing costs, enabling the use of lower resolution sensors and lowering the cost of videoscopes.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Object-affected harmful factors
If conventional denoising and sharpening techniques are used, then noise is reduced, but image sharpness deteriorates
Solution Approach 1:
The patent combines denoising and sharpening operations into a single integrated neural network (SDDTT) that processes images in one pass. This merging of previously separate processing steps allows the system to reduce noise while preserving or enhancing sharpness, resolving the contradiction between noise reduction and sharpness maintenance.
Solution Approach 2:
The neural network dynamically adjusts processing parameters based on local image characteristics, applying different levels of denoising and sharpening to different regions. This parameter adaptation allows optimal noise reduction while maintaining sharpness in critical areas, overcoming the limitations of fixed-parameter conventional techniques.
2Manufacturing precision
If higher resolution image sensors are used, then image quality improves, but device cost increases
Solution Approach 1:
The patent uses a neural network to create a computational copy or enhancement of the captured image data. Rather than relying solely on higher resolution sensors, the SDDTT processes the captured images to generate enhanced output that approaches or exceeds the quality of higher resolution sensors, providing a cost-effective alternative to hardware upgrades.
Solution Approach 2:
The patent replaces the mechanical/optical approach of using higher resolution physical sensors with a computational approach using neural network processing. This substitution of computational methods for physical hardware improvements achieves similar or better image quality at lower cost.
3Manufacturing precision
If multiple processing steps are used, then image quality improves, but processing time increases
Solution Approach 1:
The patent merges multiple processing steps (denoising, sharpening, and enhancement) into a single unified neural network processing pass. This consolidation maintains comprehensive image quality improvement while significantly reducing processing time compared to sequential application of separate processing steps.
Solution Approach 2:
The neural network performs all necessary processing operations continuously in a single pass through the network, rather than interrupting for multiple separate processing stages. This continuous processing maintains image quality while minimizing processing time and latency.
Data Source
Figure 1
Figure 2~3
Figure 4
AI summary
A method to generate near real-view images and an image processor configured to execute the method. The method includes, by an image processing circuit (100) connected to a videoscope (20'): receiving an image (202a) from the videoscope (20'); processing a source image (204a) corresponding to the image (202a) with a single denoising and feature/edge detection trained network (SDDTT) (124), the SDDTT outputting, in a single pass, a denoise map (206a) and an edge map (206b); denoising the source image (204a) with the noise map (206a) to produce a denoised image (208a); gamma-correcting the denoised image to produce a gamma-corrected image (210a); and sharpening the gamma-corrected denoised image with the edge map (208b) to produce the near real-view image (212a).