Intra-Prediction Using Artificial Neural Network for Image Encoding
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
There is a need for improved image encoding/decoding technologies that can handle high-resolution and high-definition images, particularly in the context of UHD TVs, utilizing artificial neural networks for enhanced intra-prediction modes.
Innovation Solution
An apparatus and method for image encoding/decoding using an artificial neural network to derive intra-prediction modes for target blocks, utilizing a Most Probable Mode (MPM) list and considering adjacent block information, with the neural network employing linear and nonlinear activation functions.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If conventional intra-prediction technology is used, then the encoding process is simple, but the prediction accuracy for high-resolution images is insufficient
Solution Approach 1:
The patent replaces conventional mechanical/intradiction-based prediction systems with an artificial neural network-based system. The neural network learns complex patterns from training data and automatically determines optimal prediction parameters, substituting traditional hand-crafted prediction algorithms with data-driven models that can adapt to various image characteristics and achieve superior prediction accuracy.
Solution Approach 2:
The patent changes the prediction parameters from fixed conventional values to dynamically learned parameters through neural network training. The neural network adjusts prediction parameters based on input image characteristics, allowing the system to adapt to different resolution levels and image contents, thereby improving prediction accuracy without increasing operational complexity.
2Measurement precision
If artificial neural network is introduced for intra-prediction, then prediction accuracy improves, but computational complexity increases
Solution Approach 1:
The patent performs preliminary training of the artificial neural network offline using training data, during which the network learns optimal prediction parameters and weights. Once trained, the neural network can quickly make predictions during actual encoding without requiring complex real-time computations, as the heavy lifting is done during the preliminary training phase.
Solution Approach 2:
The patent creates a simplified representation of complex image patterns through the trained neural network model. The neural network learns to capture essential features and relationships from training images, creating a compressed mental model that can quickly generate accurate predictions for new images without requiring processing of all pixel details in full resolution.
3Productivity
If neural network-based prediction is used, then compression efficiency improves, but processing time increases
Solution Approach 1:
The patent performs all heavy computational work during the preliminary training phase, where the neural network learns from training data and determines optimal prediction parameters. During actual image encoding, the system only needs to input the current image to the trained network, which quickly generates predictions based on pre-learned patterns, significantly reducing real-time processing requirements.
Solution Approach 2:
The patent applies the neural network prediction to only the most critical blocks or regions of the image that require high accuracy, rather than uniformly processing the entire image. This selective application allows the system to achieve high compression efficiency for important areas while reducing overall processing time by skipping or simplifying processing for less critical regions.
Data Source
AI summary
Disclosed herein are a method, an apparatus, and a storage medium for image encoding/decoding. An intra-prediction mode for the target block is derived, and intra-prediction for the target block that uses the derived intra-prediction mode is performed. The intra-prediction mode for the target block is derived using an artificial neural network, and an MPM list for the target block is derived using information about the target block, pieces of information about blocks adjacent to the target block, and the artificial neural network. The artificial neural network outputs one or more available intra-prediction modes. Further, the artificial neural network outputs match probabilities for one or more candidate intra-prediction modes, and each of the match probabilities for the candidate intra-prediction modes indicates a probability that the corresponding candidate intra-prediction mode matches the intra-prediction mode for the target block.