Image Decoding Using Reference Region Texture Copying
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing image encoding and decoding methods, such as H.264 AVC and HEVC, face limitations in reducing bit rate for texture regions with high spatial variability, leading to inefficient encoding and decoding processes.
Innovation Solution
The method involves applying a low-quality coding mode to target regions by determining reference regions, extracting specific texture features, and changing pixel values to match those of the reference regions, using machine learning models like CNNs to enhance image quality and reduce bit usage.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If conventional image encoding methods (H.264 AVC, HEVC) are used, then encoding compatibility and broad support are maintained, but bit rate reduction for texture regions is insufficient and encoding efficiency is limited
Solution Approach 1:
The patent copies texture features from reference regions to target regions using machine learning models. Instead of encoding all texture details traditionally, the system captures essential texture characteristics from reference areas and reconstructs them in target areas, achieving efficient compression while preserving visual quality.
Solution Approach 2:
The patent transforms image data from spatial domain to frequency domain using DCT, and applies machine learning models that operate on transformed parameters rather than raw pixel values. This parameter transformation enables more efficient encoding by focusing on perceptually important features.
2Manufacturing precision
If machine learning models are used to enhance image quality, then image quality and texture fidelity improve, but computational complexity and processing time increase
Solution Approach 1:
The patent divides the image into reference regions and target regions, applying machine learning processing only where needed. By segmenting the processing scope, the system achieves quality improvement in critical areas without uniformly applying complex computations across the entire image.
Solution Approach 2:
The patent performs DCT transformation and texture feature extraction in advance before the main encoding process. This preliminary processing prepares data in a form that is more amenable to machine learning operations, reducing the computational burden during actual encoding and decoding.
3Loss of information
If low-quality coding mode is applied to reduce bit rate, then bit rate and storage requirements decrease, but image quality and texture fidelity deteriorate
Solution Approach 1:
The patent introduces machine learning models as intermediary components between the low-quality encoded data and the final reconstructed image. These models act as smart mediators that interpret compressed data and reconstruct high-quality visual output, bridging the gap between low bit rate and high quality.
Solution Approach 2:
The patent combines traditional compression techniques (DCT, quantization) with machine learning-based texture synthesis to create a hybrid encoding system. This composite approach leverages the strengths of both methods: efficient compression from traditional techniques and quality preservation from machine learning.
Data Source
Figure 1
Figure 2
Figure 3
AI summary
Provided is a method of decoding an image, the method including determining at least one reference region to be referenced by a target region in the image to which a low-quality coding mode is applied; extracting a certain type of information from the determined at least one reference region; and changing pixel values of the target region, based on the extracted type of information.