ML Artifact Removal Model for Video Compression Quality
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional video artifact removal methods, such as using video enhancement filters, are inadequate in addressing the degradation of video quality due to artifacts like blocking, ringing, and edge burr caused by high compression rates, leading to a suboptimal user experience.
Innovation Solution
An artifact removal method and apparatus based on machine learning (ML) that trains a model to predict residuals between original and compressed video frames, enhancing the encoding and compression process to prevent noticeable artifacts in the final video.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Loss of energy
If high compression rate is used for video encoding, then storage and transmission costs are reduced, but video quality degrades due to artifacts such as blocking, ringing, and edge burr
Solution Approach 1:
The artifact removal model is applied before video encoding to pre-process the video frames and remove potential artifacts. This preliminary action prevents artifacts from being encoded into the compressed video, allowing high compression rates to be used without sacrificing video quality. The model predicts and removes artifacts in advance, so that even aggressive compression does not produce noticeable blocking, ringing, or edge burr artifacts.
2Object-affected harmful factors
If conventional video enhancement filters are used for artifact removal, then some noise can be reduced, but the effect on removing compression artifacts is limited
Solution Approach 1:
The patent replaces conventional mechanical video enhancement filters with a machine learning-based artifact removal model. This model uses deep learning techniques to predict and remove compression artifacts, providing superior effectiveness compared to traditional filter-based methods. The neural network learns complex patterns of artifacts from training data and can remove them more accurately and reliably than conventional filters.
3Adaptability or versatility
If multiple types of video artifacts are addressed separately, then each artifact type can be treated specifically, but the system complexity increases
Solution Approach 1:
The artifact removal model is designed as a universal solution that can handle multiple types of compression artifacts simultaneously. The machine learning model learns to recognize and remove various artifact types (blocking, ringing, edge burr, etc.) through a single unified architecture, eliminating the need for multiple separate filters or complex filter combinations. This provides versatile artifact removal capability while maintaining system simplicity.
Data Source
AI summary
Provided are an artifact removal method and apparatus based on machine learning (ML), and a method and apparatus for training an artifact removal model based on ML, in the field of artificial intelligence (AI). In the method, artifacts that may be generated in a process of compressing a video is preprocessed and removed by using an artifact removal model having a residual learning capability.