Depth-Guided Filtering for Video Conference Noise Reduction
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing video conferencing technologies face challenges in enhancing video encoding and decoding under bitrate constraints, particularly in preserving depth details and reducing noise in depth maps, which can introduce artifacts during image segmentation and transmission.
Innovation Solution
The implementation of depth-guided filtering, which generates a filtered depth map to improve image quality by combining depth and texture data, upsampling lower-resolution depth maps, and applying filters to reduce noise and enhance detail preservation, especially for objects closer to the viewpoint, integrated into the video encoding-decoding loop.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If depth maps are used for image segmentation and transmission, then depth information is preserved, but noise and artifacts are introduced
Solution Approach 1:
The patent extracts and separates depth information from color image data into distinct depth maps. By isolating depth information into separate maps, the system can process and filter depth data independently from color data, enabling targeted noise reduction while preserving depth accuracy for segmentation tasks.
Solution Approach 2:
The patent applies different filtering strategies to different regions of the depth map based on local characteristics. Objects closer to the viewpoint receive different processing than distant objects, with the system adaptively adjusting filter strength and type to preserve important depth details while removing noise in specific local regions.
2Productivity
If video encoding is performed under bitrate constraints, then transmission efficiency is improved, but image quality deteriorates
Solution Approach 1:
The patent segments video data into separate color and depth components that can be encoded and transmitted independently. This segmentation allows the system to allocate bitrate differently across components, potentially using fewer bits for depth maps while maintaining overall perceptual quality, thus improving transmission efficiency without sacrificing image quality.
Solution Approach 2:
The patent adds a depth dimension to traditional video encoding by incorporating depth maps as a separate data layer. This dimensional expansion enables new compression strategies where depth information, being more compressible than full-color images, can carry significant scene information at lower bitrates, improving overall transmission efficiency.
3Object-affected harmful factors
If depth maps are filtered to reduce noise, then artifact reduction is achieved, but important depth details may be lost
Solution Approach 1:
The patent employs dynamic filtering where filter parameters are adjusted based on local depth map characteristics, object distance from viewpoint, and scene complexity. Rather than applying uniform filtering, the system dynamically adapts filter strength and type to preserve important depth edges and details while removing noise in appropriate regions.
Solution Approach 2:
The patent creates multiple versions or copies of depth map data at different processing stages, allowing the system to preserve original high-detail depth information while generating filtered versions for specific uses. This copying strategy enables the system to have both noisy high-fidelity depth maps for segmentation and cleaned depth maps for transmission without losing important details.
Data Source
AI summary
A method is provided in one example embodiment that includes generating a depth map that corresponds to a video image and filtering the depth map with the video image to create a filtered depth map. The video image can be filtered with the filtered depth map to create an image. In one example implementation, the video image is filtered using extended depth-guided filtering that is incorporated into a video encoding-decoding loop.