Image Shadow Removal Using Synthetic Training Data
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Captured images often include shadows from obstructions, which affect the imaging effect and user experience, necessitating an effective method for shadow removal.
Innovation Solution
An image processing method and apparatus that acquires a target image with a shadow and removes it using a trained target model. The target model is trained on target image samples with shadows and corresponding shadow image samples, generated from captured shadow-free images and emulated imaging conditions.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Manufacturing precision
If shadow removal processing is performed on captured images, then imaging quality and clarity are improved, but processing complexity and computational resources increase
Solution Approach 1:
The system performs preliminary actions by capturing a shadow-free reference image simultaneously with the shadowed target image. This reference image is used later in the shadow removal process to provide clean illumination information, avoiding the need for complex real-time shadow analysis and simplifying the overall processing while maintaining high imaging quality.
Solution Approach 2:
The patent introduces an intermediary approach by using a reference image as a mediator between the shadowed target image and the final shadow-removed output. The reference image serves as an intermediate component that provides clean illumination data, which is then combined with the target image through pixel-level operations to remove shadows efficiently without requiring complex computational models.
2Measurement precision
If multiple image samples and emulated imaging conditions are used for model training, then shadow removal accuracy is improved, but data processing time and computational cost increase
Solution Approach 1:
The system creates synthetic shadowed images by copying and overlaying shadow patterns from reference images onto shadow-free target images. This copying approach generates large amounts of training data with consistent ground truth labels, improving model accuracy while avoiding the time-consuming process of manually annotating real shadowed images or capturing diverse real-world shadow scenarios.
Solution Approach 2:
The patent applies parameter changes by systematically varying imaging conditions (lighting angles, obstruction positions, shadow intensities) in the synthetic data generation process. By changing these parameters programmatically, the system generates diverse training samples that improve model robustness and accuracy without requiring proportional increases in manual processing time or computational resources.
Data Source
AI summary
An image processing method and apparatus belong to the technical field of communication. The method comprises: acquiring a target image including a shadow; and removing the shadow from the target image based on a target model, to acquire an image after shadow removal, wherein the target model is a model trained based on a target image sample including a shadow and a shadow image sample corresponding to the target image sample, the target image sample is a sample generated based on a captured shadow-free image sample and a preset emulated imaging condition, and the shadow image sample is a sample determined based on the target image sample and the shadow-free image sample.