Label-Image-Guided Block Division for Low-Resource Image Encoding
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional image encoding techniques require significant resource consumption due to the use of trained neural networks and pre-processing for determining encoding blocks, leading to increased circuit scale and cost.
Innovation Solution
An image encoding device that determines a division pattern for encoding blocks based on a label image, where labels represent feature types, allowing for efficient block division without excessive resource consumption.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If trained neural networks and pre-processing are used to determine encoding blocks, then encoding accuracy is improved, but resource consumption and circuit scale increase
Solution Approach 1:
The patent extracts only the essential feature information needed for block division from the image, rather than using complete neural network processing. By taking out and utilizing only the necessary feature amounts (such as edge information, texture information, or object detection results), the system achieves accurate block division while avoiding the complexity of full neural network implementations.
Solution Approach 2:
The patent uses pre-generated label images or feature maps as simplified copies of the full image data, which already contain the necessary feature information for block division. These copied feature representations eliminate the need for real-time neural network processing during encoding, reducing circuit requirements while maintaining encoding accuracy.
2Manufacturing precision
If trained neural networks are used to determine encoding blocks, then encoding quality is improved, but cost and resource consumption increase
Solution Approach 1:
The patent performs feature extraction and label generation as preliminary actions before the actual encoding process. By preparing label images or feature maps in advance (either through separate processing or pre-computation), the system eliminates the need for resource-intensive neural network operations during real-time encoding, thus maintaining quality while reducing resource consumption.
Solution Approach 2:
The patent employs lightweight, temporary data structures and feature representations (such as simple label maps or feature flags) that can be generated quickly and discarded after use, replacing the need for persistent, resource-heavy neural network models. These simple feature objects provide sufficient information for block division without requiring substantial computational resources.
3Measurement precision
If feature extraction using machine learning is used to determine block division patterns, then encoding precision is improved, but device complexity and resource usage increase
Solution Approach 1:
The patent segments the image processing task into distinct stages: feature extraction (using simple methods), label generation, and block division determination. By separating these functions and using simplified feature extraction methods (such as traditional image processing algorithms or pre-computed features), the system achieves precise block division without requiring complex integrated machine learning systems.
Solution Approach 2:
The patent introduces label images or feature maps as intermediary data structures that bridge the gap between simple image input and complex block division decisions. These intermediaries contain pre-processed feature information that guides block division without requiring the encoder to perform complex machine learning operations, thus reducing processing complexity while maintaining precision.
Data Source
AI summary
An image encoding device includes: an image acquisition unit that acquires a label image in which a label representing a type of a feature is assigned to a region of the label image corresponding to the feature in a label target image; a block division determination unit that determines a division pattern of an encoding target block in an encoding target image based on the region in the label image and the label assigned to the region; and an encoding unit that executes an encoding process for the encoding target block specified by the division pattern.