Hybrid Video Compression With ML Frame Interpolation and DCT Residuals
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing video compression standards, such as HEVC/H.265, require significant computing power and face limitations in achieving further compression improvements, while advanced video synthesis techniques struggle with non-linear motion and occlusions, especially with increased anchor-frame distances.
Innovation Solution
A hybrid approach combining DCT-based video compression with machine learning-based frame interpolation, where intermediate frames are synthesized using deep learning models and selectively encoded residual information to improve areas with complex motion or occlusions, resulting in hybrid frames that reduce bandwidth and maintain image quality.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Quantity of substance
If DCT-based video compression techniques are used to achieve higher compression ratios, then bitrate is reduced, but image quality deteriorates and computing power requirements increase
Solution Approach 1:
The video compression process is segmented into two distinct parts: traditional DCT-based compression for anchor frames and machine learning-based synthesis for intermediate frames. This segmentation allows each method to operate in its optimal domain, with DCT handling structured compression and ML handling complex motion interpolation, thereby maintaining image quality while reducing bitrate
Solution Approach 2:
The patent creates a composite compression system that combines traditional DCT-based video coding with machine learning-based frame synthesis. This hybrid approach integrates the strengths of both methods: DCT provides efficient baseline compression while ML adds intelligent synthesis capabilities for intermediate frames, achieving superior compression ratios without sacrificing image quality
2Productivity
If DCT-based video compression standards are advanced to achieve better compression, then compression efficiency improves, but computing power requirements increase tenfold
Solution Approach 1:
Instead of applying computationally intensive processing to all frames, the patent applies machine learning synthesis only to intermediate frames while using efficient DCT compression for anchor frames. This partial application of advanced techniques reduces overall computing power requirements while maintaining compression efficiency
Solution Approach 2:
The patent introduces machine learning-based frame synthesis as an intermediary method between traditional compression standards. This intermediary approach handles the complex motion and occlusion cases that traditional DCT-based methods struggle with, improving compression efficiency without requiring proportional increases in computing power
3Ease of manufacture
If traditional DCT-based compression is used to handle non-linear motion and occlusions, then implementation is simpler, but image quality deteriorates
Solution Approach 1:
The patent substitutes traditional mechanical DCT-based motion compensation with machine learning-based frame synthesis for handling non-linear motion and occlusions. The ML model learns complex motion patterns and synthesizes intermediate frames more accurately, improving image quality while the system maintains simplicity through automated learning rather than complex rule-based implementations
Data Source
AI summary
Systems and methods for hybrid video compression. A method includes receiving an encoded video including at least two compressed frames corresponding to at least two anchor frames and a compressed subset of at least one intermediate frame between the at least two anchor frames, generating, by inputting the at least two anchor frames into a deep machine learning model, a synthesized image frame corresponding to the at least one intermediate frame between the at least two anchor frames, reconstructing at least one hybrid image frame by combining the compressed subset of the at least one intermediate frame with the synthesized image frame, and outputting a video including the at least two anchor frames and the at least one hybrid image frame. Thus, DCT-based video compression techniques can leverage machine learning-based video interpolation techniques to provide encoded video streams with a reduced bitrate and thus reduced bandwidth while maintaining image quality.