Low-Bitrate Video Content Generation With ML Image Restoration
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing content distribution systems face challenges in reducing the load on communication pathways while maintaining reasonable image quality for large-volume video content transmission, as existing methods either compromise image quality or do not address the load on communication configurations.
Innovation Solution
A content generating device and system that utilizes low-bit-rate encoded data and machine-learned model data to enhance image quality, including meta-information like QP, prediction error coefficients, and motion vectors, and employs a machine-learned model to improve perceptual quality through deep learning.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If the transmission rate (bit rate) is reduced to compress bandwidth for video image content, then the data volume is reduced and transmission efficiency is improved, but the image quality deteriorates with loss of detailed information and appearance of block noise or mosquito noise
Solution Approach 1:
The video stream is divided into multiple scenes, and for each scene, a scene type is determined and video encoding parameters specific to that scene type are applied. This segmentation allows different parts of the video to be encoded with appropriate parameters, improving overall compression efficiency while maintaining quality where needed.
Solution Approach 2:
Different video encoding parameters are applied to different scene types locally. For example, scenes with high-speed motion use parameters optimized for motion compression, while still scenes use parameters optimized for detail preservation. This local adaptation resolves the contradiction by applying appropriate quality levels to different regions of the video content.
2Manufacturing precision
If high-resolution cameras and display devices are used to improve image quality, then the visual fidelity is enhanced, but the data volume of image content increases significantly
Solution Approach 1:
The system dynamically changes video encoding parameters based on scene type detection. By identifying characteristics of each scene (such as motion intensity, complexity, and content type), the encoder adjusts parameters like quantization step size, transformation type, and prediction mode to achieve optimal compression ratios while maintaining perceptual quality.
3Productivity
If video encoding parameters are defined for each scene type to improve compression efficiency, then the transmission load is reduced, but images that do not fall within any predefined scene types may not be effectively encoded
Solution Approach 1:
The scene type determination mechanism is designed to be universal, covering a comprehensive set of scene types that can accommodate various video content. The system includes default handling for unrecognized scenes and can adapt to new scene types by extending the classification framework, ensuring both high compression efficiency for known types and reasonable performance for unknown types.
Data Source
AI summary
According to one or more embodiments, a content generating device is provided. The content generating device comprises a first generator, a second generator, and a first transmitter. The first generator generates low-bit-rate encoded data that is original data having been encoded to a low-bit-rate. The second generator generates machine-learned model data for generating improved data obtained by improving human perceptually the low-bit-rate encoded data, based on a machine-learned model. The first transmitter transmits the low-bit-rate encoded data and the model data to outside.