AI Image Coding With Neural Up-Scaling for Low-Bitrate Quality
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing image encoding and decoding technologies struggle to efficiently process high-resolution/high-quality images at low bitrates, leading to potential information loss during compression and decompression.
Innovation Solution
An AI-based encoding and decoding method and apparatus that utilizes jointly trained deep neural networks for down-scaling and up-scaling images, reducing the bitrate by first down-scaling high-resolution images, encoding them at a lower resolution, and then up-scaling them back to the original resolution during decoding, while maintaining image quality.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Loss of energy
If high-resolution images are encoded directly at low bitrate, then storage and transmission efficiency is improved, but image quality and information completeness deteriorate
Solution Approach 1:
The patent applies preliminary action by down-scaling the high-resolution image before encoding. The image is first processed through a down-scaling neural network to reduce its resolution, and only then is it encoded at low bitrate. This preliminary resolution reduction enables efficient low-bitrate encoding while preserving quality through subsequent up-scaling.
Solution Approach 2:
The patent introduces an intermediary approach by using neural networks as mediators between the high-resolution original image and the low-bitrate encoded version. The down-scaling neural network acts as an intermediary that processes the image before encoding, and the up-scaling neural network acts as another intermediary that restores the image after decoding, thereby enabling quality preservation at low bitrate.
2Productivity
If image resolution is reduced for encoding, then encoding efficiency and bitrate are improved, but image detail and fidelity deteriorate
Solution Approach 1:
The patent applies preliminary action by performing down-scaling before encoding. The image is first processed through a down-scaling neural network to reduce resolution, which enables more efficient encoding at lower bitrate. The preliminary resolution reduction is followed by up-scaling after decoding to restore image fidelity.
Solution Approach 2:
The patent applies parameter changes by dynamically adjusting the resolution parameter of the image. The down-scaling neural network changes the resolution parameter to a lower value for encoding, and the up-scaling neural network changes it back to the original resolution for output. This parameter manipulation enables both encoding efficiency and image fidelity.
3Device complexity
If conventional encoding is used without AI processing, then processing simplicity is maintained, but quality loss and information omission occur
Solution Approach 1:
The patent applies mechanics substitution by replacing conventional mechanical encoding processes with AI-based neural network processing. Instead of using traditional compression algorithms, the patent employs down-scaling and up-scaling neural networks that can intelligently preserve image quality while enabling low-bitrate encoding, thereby reducing information loss.
Solution Approach 2:
The patent applies parameter changes by introducing AI-based parameter transformation through neural networks. The down-scaling and up-scaling neural networks dynamically adjust image parameters (resolution, quality) in a way that conventional methods cannot, enabling quality preservation even at low bitrates where conventional encoding would fail.
Data Source
Figure 1
Figure 2
Figure 3
AI summary
Provided is an artificial intelligence (Al) decoding apparatus including: a memory; and a processor, wherein the processor is configured to obtain AI data related to AI down-scaling of an original image, and image data generated as a result of encoding a first image, obtain a second image corresponding to the first image by decoding the image data, determine a resolution ratio in a horizontal direction and a resolution ratio in a vertical direction, between the original image and the first image, based on the AI data, and obtain, via an up-scaling deep neural network (DNN), a third image, in which a resolution in at least one of a horizontal direction and a vertical direction is increased from the second image, wherein the resolution ratio in the horizontal direction and the resolution ratio in the vertical direction are determined as different values.