Image Decomposition for Lighting Adaptation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current e-commerce websites struggle to accurately represent product appearances under different lighting conditions, as images are typically captured under specific lighting and fail to reflect how products appear in various environments, leading to inadequate user assessment.
Innovation Solution
An image generation system decomposes input images into shading and reflectance components, using a machine learning model to generate output images under specified lighting conditions, allowing for accurate representation of products in different lighting scenarios.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If product images are captured under specific lighting conditions, then the images can be clearly captured with good quality, but the images fail to represent product appearance under different lighting conditions
Solution Approach 1:
The patent segments the product image into multiple components including reflectance map, shading map, and normal map. This segmentation allows the system to separately process and manipulate different aspects of the image, enabling the reflectance component to be preserved while the shading component is modified to represent different lighting conditions.
Solution Approach 2:
The patent changes the lighting parameters in the shading map while keeping the reflectance map constant. By adjusting parameters such as light source position, direction, and intensity in the shading component, the system generates images representing different lighting conditions without altering the intrinsic reflectance properties of the product.
2Adaptability or versatility
If multiple images are captured under different lighting conditions, then product appearance under various lighting can be represented, but the complexity and cost of image acquisition increases
Solution Approach 1:
Instead of capturing multiple physical images under different lighting conditions, the patent creates synthetic copies by computationally generating new shading maps based on the original image decomposition. This copying approach allows infinite variations of lighting conditions to be generated from a single captured image, eliminating the need for complex multi-lighting acquisition setups.
Solution Approach 2:
The patent replaces the mechanical approach of physically changing lighting conditions with multiple cameras or light sources with a computational approach. By using image decomposition and rendering techniques, the system substitutes physical lighting manipulation with algorithmic shading map generation, significantly reducing device complexity.
3Ease of manufacture
If traditional image processing is used, then the processing method is simple, but it cannot accurately separate lighting effects from object properties
Solution Approach 1:
The patent performs preliminary decomposition of the image into reflectance and shading components before any lighting transformation. This preliminary action of separating the intrinsic object properties (reflectance) from the extrinsic lighting effects (shading) enables accurate manipulation of lighting conditions while preserving object characteristics, avoiding the need for complex post-processing corrections.
Data Source
AI summary
An image generation system generates images of objects under different lighting conditions. An image of an object and lighting conditions for an output image are received. The lighting conditions may specify, for instance, a location and/or color of one or more light sources. The image of the object is decomposed into a shading component and a reflectance component. A machine learning model takes the reflectance component and specified lighting conditions as input, and generates an output image of the object under the specified lighting conditions. In some configurations, the machine learning model may be trained on images of objects labeled with object classes, and the output image may be generated by also providing an object class of the object in the image as input to the machine learning model.