Generalized Difference Coder for Residual Coding in Video Compression

Simple SummaryContent extracted from patent full text and abstract with AI.

This patent describes a novel method and system for video and image compression that leverages neural networks to improve how differences (residuals) between predicted and actual video frames are encoded and decoded. The invention introduces the concept of a "generalized difference coder" that uses a neural network to extract advanced features (generalized residuals) between the predicted and original signals during encoding, which are then efficiently compressed and stored. During decoding, another neural network reconstructs the original signal from these features and the prediction, potentially improving compression efficiency and image quality compared to traditional linear residual coding methods.

Use CasesContent extracted from patent full text and abstract with AI.

• Video streaming services (e.g., Netflix, YouTube) to achieve higher compression rates and better visual quality at lower bitrates.
• Video conferencing and real-time communication (e.g., Zoom, Teams) to reduce required bandwidth while maintaining image fidelity.
• Mobile video recording and playback to save storage space and prolong battery life.
• Surveillance camera systems for more efficient storage and transmission of large volumes of security footage.
• Cloud gaming and remote rendering services that require efficient low-latency video transmission.
• Archiving and distributing medical imaging, satellite imagery, or other high-resolution visual data.
• Implementation in video codecs for television broadcasts, Blu-ray discs, and digital cinema distribution.

BenefitsContent extracted from patent full text and abstract with AI.

• Improved compression efficiency leading to smaller file sizes and less bandwidth usage for comparable or better visual quality.
• Enhanced adaptability through AI-based neural networks that learn complex patterns and redundancies beyond traditional methods.
• Potential for lower latency and better robustness in real-time applications due to more efficient residual coding.
• Flexibility to be implemented in hardware or software, supporting a range of devices from mobile phones to dedicated servers.
• Compatibility with existing video standard frameworks (e.g., H.264, HEVC, VVC) or as enhancements/extensions to current codecs.
• Noise reduction and more effective handling of "skip areas" (regions with little to no change) through intelligent default value assignment.
• Scalability and potential for further improvement as neural network models and training methods evolve.

Technical Classifications (CPCs)

Inventors & Applicants

Patent Abstract

This application provides methods and apparatuses for encoding image or video related data into a bitstream. The present disclosure may be applied in the field of artificial intelligence (Al)-based video or picture compression technologies, and in particular, to the field of neural network-based video compression technologies. A neural network (generalized difference) is applied to a signal and a predicted signal during the encoding to obtain a generalized residual. During the decoding another neural network (generalized sum) may be applied to a reconstructed generalized residual and the predicted signal to obtain a reconstructed signal.