Multi-Stage Image Deblurring Using ML Feedback
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing image deblurring techniques face challenges in accurately recovering blur features from single blurred images due to loss of details, leading to inefficient computation and resource consumption, particularly in low-light and underwater scenarios.
Innovation Solution
A method and electronic device that employ multi-stage encoding and decoding processes using point-wise and depth-wise convolutions to extract coarser and finer blur features, with machine learning models for threshold-based feature map generation and deconvolution, reducing computational and storage resources while enhancing deblurring performance.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If traditional deconvolution operations are used to deblur images, then deblurring can be achieved, but accurate knowledge of blur features is required which is difficult to recover from single blurred images
Solution Approach 1:
The patent segments the blur feature extraction process into multiple stages: initial blur feature estimation, multi-scale feature map generation, and iterative refinement. Each stage processes different aspects of blur characteristics, breaking down the difficult task of accurate blur feature recovery from single blurred images into manageable steps that progressively improve precision.
Solution Approach 2:
The patent introduces multi-scale feature maps that operate at different spatial resolutions and dimensional levels. By processing blur features across multiple scales and dimensions rather than a single representation, the system overcomes the limitation of insufficient information in single blurred images and achieves accurate blur feature recovery.
2Manufacturing precision
If detailed processing is performed to recover blur features accurately, then deblurring quality improves, but computation and memory requirements increase
Solution Approach 1:
The processing pipeline is segmented into distinct modules: feature extraction, multi-scale feature map generation, blur feature estimation, and image reconstruction. Each module handles specific computations with optimized resource usage, allowing high deblurring quality without requiring uniformly high computational complexity throughout the entire system.
Solution Approach 2:
The patent applies partial processing by generating feature maps at multiple scales only where needed and iteratively refining blur features to the necessary degree of accuracy. This avoids performing excessive computations that would be required for perfect reconstruction, achieving satisfactory deblurring quality with reduced computational and memory requirements.
3Productivity
If multi-stage encoding and decoding is implemented, then resource consumption is reduced, but processing time may increase
Solution Approach 1:
The patent performs preliminary encoding of the blurred image into multi-scale feature maps before the actual deblurring computation. This preliminary organization of image data into structured feature representations at different scales enables more efficient processing during the decoding stage, reducing overall resource consumption while maintaining acceptable processing time through optimized data organization.
Data Source
AI summary
A method for deblurring a blurred image includes encoding, by at least one processor, a blurred image at a plurality of stages of encoding to obtain an encoded image at each of the plurality of stages; decoding, by the at least one processor, an encoded image obtained from a final stage of the plurality of stages of encoding by using an encoding feedback from each of the plurality of stages and a machine learning (ML) feedback from at least one ML model; and generating, by the at least one processor, a deblurred image in which at least one portion of the blurred image is deblurred based on a result of the decoding.