High-Resolution Image Shadow Removal with Two-Stage AI
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing shadow removal techniques for High Resolution (HR) images are inefficient, require high computational power, and are not feasible for resource-constrained platforms like smartphones, often leading to image quality loss and reliance on manual intervention.
Innovation Solution
A method involving conversion of HR images to Low Resolution (LR) using a first AI model, followed by generating an LR shadow-free image and then up-scaling it to HR using a second AI model, leveraging lightweight network architectures to retain image quality and reduce computational resources.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If existing shadow removal techniques process HR images directly, then shadow removal can be performed, but high computational power is required and image quality is degraded
Solution Approach 1:
The patent segments the shadow removal process into two distinct stages: first converting HR image to LR image for shadow removal, then converting the shadow-free LR image back to HR image. This segmentation allows each stage to be optimized independently, reducing overall computational requirements while maintaining image quality.
Solution Approach 2:
The patent introduces an intermediate LR shadow-free image as a mediator between the input HR shadowed image and the final HR shadow-free image. This intermediate representation enables shadow removal to be performed on a smaller, computationally cheaper image while preserving the ability to reconstruct high-quality output.
2Power
If HR images are converted to LR images for shadow removal, then computational resources are reduced, but image quality is lost
Solution Approach 1:
The patent performs preliminary shadow removal on the LR image before converting back to HR. By removing shadows in the LR domain first, the problematic shadow artifacts are eliminated before the upscaling process, preventing quality degradation in the final HR image.
Solution Approach 2:
The patent changes the resolution parameter from HR to LR during the shadow removal phase to reduce computational complexity, then changes it back to HR in the reconstruction phase. This dynamic parameter adjustment allows the system to benefit from both low-resolution processing efficiency and high-resolution output quality.
3Measurement precision
If manual intervention is used to remove shadows, then shadow removal can be performed accurately, but the process is time-consuming and requires human resources
Solution Approach 1:
The patent implements an automated system that performs shadow removal without requiring manual intervention. The AI model automatically detects shadows, determines shadow-free regions, and reconstructs the image, enabling the system to serve itself and eliminating the need for human operators.
Solution Approach 2:
The patent replaces the mechanical process of manual shadow removal with an automated AI-based system. The neural network model automatically identifies shadow regions and performs removal operations, substituting human manual work with computational algorithms that are both accurate and efficient.
4Productivity
If heavy network architectures are used for shadow removal, then processing capability is improved, but deployment on resource-constrained platforms becomes infeasible
Solution Approach 1:
The patent applies local quality by performing shadow removal operations specifically on shadow-affected regions rather than processing the entire HR image at full resolution. This localized approach reduces the computational burden on the network architecture, enabling deployment on resource-constrained platforms while maintaining effective shadow removal capability.
Data Source
AI summary
The present disclosure relates to method and apparatus for removing shadows from a High Resolution (HR) image. The HR image including shadows is received. The HR image converted into Low Resolution (LR) image. LR shadow-free image is generated from the LR image using first Artificial Intelligence (AI) model. A HR shadow-free image is generated based on combination of the LR shadow-free image and the HR image using second AI model. The present disclosure provides an efficient framework and lightweight network architectures which are used to remove the shadows based on corresponding shadow characteristics. The framework can be deployed in resource-constrained platforms. The present disclosure facilitates processing selected areas of the HR image and retaining the unselected area. Thus, less computational resources are required.