Single Image Reflection Removal via Edge Map Prediction
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional systems for removing reflections from single digital images are ineffective, especially in high-contrast images with multiple reflections, as they rely on limited information and struggle to distinguish between reflections and intended objects.
Innovation Solution
A computing device implements a removal system that processes a digital image depicting light reflected by a surface and light transmitted through the surface, using machine learning models to predict an edge map of the transmitted image and a reflected component, thereby generating a corrected digital image without reflected light.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If conventional systems use multiple captured images to remove reflections, then reflection removal accuracy is improved, but storage capacity consumption increases and processing burden increases
Solution Approach 1:
The patent extracts only the essential information (edge maps and reflected component predictions) from the digital image using machine learning models, rather than processing the complete original image data. This allows reflection removal to be performed on a single image while consuming minimal storage capacity, as only the processed feature data is retained rather than the full original image.
Solution Approach 2:
The patent creates a simplified representation (copy) of the image data in the form of edge maps and reflected component predictions. These copies contain the essential information needed for reflection removal but occupy significantly less storage space than the original image, enabling accurate reflection removal without consuming excessive storage capacity.
2Quantity of substance
If conventional systems use limited information from a single digital image to remove reflections, then storage capacity consumption is reduced, but reflection removal accuracy deteriorates
Solution Approach 1:
The patent replaces conventional mechanical/optical approaches (multiple image capture) with machine learning-based information processing. By using trained machine learning models to extract edge maps and reflected components from single image data, the system achieves high-accuracy reflection removal while consuming minimal storage capacity, as the models process and interpret the limited information efficiently.
Solution Approach 2:
The patent transforms the image data into different parameter representations (edge maps and reflected component predictions) through machine learning processing. This parameter transformation allows the system to work with reduced information content while maintaining or improving reflection removal accuracy, as the transformed parameters capture the essential reflective characteristics more effectively than raw pixel data.
3Measurement precision
If conventional systems process multiple captured images to identify and remove reflections, then reflection removal completeness is improved, but processing time and computational burden increase
Solution Approach 1:
The patent performs preliminary processing by pre-training machine learning models on reflection removal tasks. These pre-trained models can quickly process single images and generate edge maps and reflected component predictions without requiring time-consuming multi-image capture and processing sequences, thus reducing processing time while maintaining completeness.
Solution Approach 2:
The patent creates simplified copies of the image data in the form of edge maps and reflected component predictions, which can be processed much faster than the original complete image data. These copies retain the essential reflection information needed for complete removal but require significantly less computational processing time, enabling fast and complete reflection removal.
4Ease of operation
If conventional systems rely on limited information from a single digital image, then ease of operation is improved, but ability to distinguish between reflections and objects deteriorates
Solution Approach 1:
The patent replaces simple single-image processing with machine learning-based analysis that automatically distinguishes reflections from objects. The trained models analyze edge maps and reflected components to identify reflective characteristics, enabling accurate distinction between reflections and objects while maintaining ease of operation through automated processing.
Solution Approach 2:
The patent transforms the visual characteristics of the image data into different representations (edge maps with edge information and reflected component predictions). This transformation highlights the reflective properties of light and enables clear distinction between reflections and objects, as the processed data emphasizes the characteristics that differentiate these elements.
Data Source
AI summary
In implementations of systems for single image reflection removal, a computing device implements a removal system to receive data describing a digital image that depicts light reflected by a surface and light transmitted through the surface. The removal system predicts an edge map of a transmitted image for the light transmitted through the surface by processing the data using a first machine learning model trained on a first type of training data. A reflected component is predicted for the light reflected by the surface by processing the data using a second machine learning model trained on a second type of training data. A corrected digital image is generated that does not depict the light reflected by the surface based on the data, the edge map of the transmitted image, and the reflected component.