Adaptive Image Preprocessing Kernel for Hybrid Vision Optimization
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional video coding technologies are optimized for human perceptual quality and fail to adapt to the changing environment where images are consumed by both humans and machines, requiring a system that can dynamically optimize between perceptual quality and task execution performance.
Innovation Solution
A hybrid vision system that uses adaptive image preprocessing and reconstruction methods, employing a preprocessing kernel and reconstruction network adjustable by a control parameter to balance human vision and machine vision needs, enabling selective optimization of bitstream and image quality.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Manufacturing precision
If video coding technology is optimized for human perceptual quality, then subjective image quality and objective image quality indexes are improved, but task execution performance for machines deteriorates
Solution Approach 1:
The patent implements dynamic optimization by introducing a control parameter that allows the system to adaptively switch between human perceptual quality optimization and machine task execution performance optimization. The preprocessing kernel and reconstruction network are dynamically adjusted based on the control parameter value, enabling the system to respond to different application requirements in real-time.
Solution Approach 2:
The patent changes the optimization parameter from fixed (either human perceptual quality or machine task performance) to variable through the control parameter. By modifying the control parameter value, the system can shift the optimization focus between perceptual quality metrics (PSNR, SSIM) and machine vision task performance, resolving the contradiction through parameter variability.
2Productivity
If images are irreversibly transformed to improve machine task execution performance, then task execution performance is improved, but image reconstruction capability for humans deteriorates
Solution Approach 1:
The patent applies preliminary action by performing reversible preprocessing operations before encoding. The preprocessing kernel modifies the input image in a controlled manner that preserves reconstruction capability, and the reconstruction network later reverses these transformations. This preliminary reversible transformation enables both machine task execution and human image reconstruction.
Solution Approach 2:
The patent implements discarding and recovering by temporarily transforming image characteristics during preprocessing to improve machine task execution, then recovering the original image characteristics through the reconstruction network. The control parameter governs the extent of transformation and recovery, ensuring that machine performance benefits are achieved without permanently sacrificing reconstruction capability.
3Adaptability or versatility
If a hybrid vision system supports both human vision and machine vision, then system versatility is improved, but system complexity increases
Solution Approach 1:
The patent achieves universality by designing a single hybrid vision system that performs both human perceptual quality optimization and machine task execution performance optimization. Through the control parameter and adjustable preprocessing/reconstruction components, one system fulfills multiple functions that would otherwise require separate systems, managing complexity while maintaining versatility.
4Adaptability or versatility
If a system dynamically generates bitstream and reconstructed image optimized for task or image quality, then system adaptability is improved, but processing complexity increases
Solution Approach 1:
The patent implements dynamic generation through the control parameter that governs the preprocessing kernel and reconstruction network configuration. Based on the control parameter value, the system dynamically adjusts the preprocessing intensity and reconstruction strategy, enabling adaptive optimization between task performance and image quality without requiring multiple fixed systems.
Data Source
AI summary
Disclosed herein is a method for adaptive image preprocessing and reconstruction. The method includes preprocessing an input image, encoding and decoding the preprocessed image, and reconstructing the encoded and decoded image. Here, preprocessing the input image may be performed using a preprocessing kernel generated based on a control parameter indicating a weight for human vision and machine vision.