Intrinsic Image Decomposition Using Constrained Shadow Color Space
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing intrinsic image decomposition methods struggle to separate an input image into an object color image and a shadow image with high precision, as the problem is inherently ill-posed and previous techniques assume known object colors or grayscale shadows, limiting their accuracy.
Innovation Solution
An image processing apparatus and method that estimates an object color image and shadow image by limiting the color space of the shadow component to a predetermined condition, using a convolutional neural network (CNN) to process input images and constrain the shadow estimation within a defined color space, such as black-body radiation or CIE daylight models, or based on lighting conditions like direct and global light.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If intrinsic image decomposition is performed without constraining the color space of shadow components, then the method can handle general cases, but the separation precision into object color and shadow images deteriorates
Solution Approach 1:
The patent applies parameter changes by constraining the color space of shadow components to specific color spaces (black-body radiation locus, CIE daylight locus, or user-defined color spaces) based on predetermined color conditions. This parameter constraint transforms the ill-posed decomposition problem into a well-posed one, enabling high-precision separation while maintaining adaptability through multiple selectable color space models.
Solution Approach 2:
The patent implements dynamics by allowing flexible selection of different color space models (black-body radiation, CIE daylight, or custom color spaces) based on the imaging conditions. The system can dynamically adapt to different lighting scenarios while maintaining precision through the chosen color space constraints.
2Measurement precision
If the shadow image is estimated without color space constraints, then the method is simpler to implement, but the accuracy of shadow estimation deteriorates
Solution Approach 1:
The patent improves shadow estimation accuracy by introducing color space parameter constraints. The shadow estimation unit limits the color space of shadow components to predetermined color spaces, which guides the decomposition process and significantly improves accuracy despite the increased implementation complexity of defining and enforcing these color space boundaries.
3Ease of manufacture
If object color is assumed to be known, then the decomposition problem becomes easier to solve, but the method loses versatility for unknown objects
Solution Approach 1:
The patent resolves this contradiction by changing the approach from assuming known object colors to constraining shadow color parameters instead. By limiting the color space of shadow components to predetermined color spaces, the system makes the decomposition problem solvable without prior knowledge of object colors, thereby maintaining versatility for unknown objects while keeping the decomposition process tractable.
Data Source
AI summary
To enable separation of an input image into an object color image and a shadow image with a high precision. An image processing apparatus includes an object color estimation unit that estimates an object color image having a color component of an object included in an input image as a pixel value on the basis of a feature amount of the input image, and a shadow estimation unit that estimates a shadow image having a shadow component of the input image as a pixel value on the basis of the feature amount of the input image. The shadow estimation unit estimates the shadow image by limiting a color space that can be taken by the shadow component of the input image to a color space determined under a predetermined color condition. The technology of the present disclosure can be applied to, for example, an image processing apparatus or the like that separates an input image into an object color image and a shadow image.