Sparse Correspondence Map for Object Appearance Transfer
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional image processing systems fail to accurately perform object appearance transfer when objects in the content image and style image have complex geometries and extensive geometry variations, resulting in poor capture of detailed textures from the style image while retaining the spatial structure from the content image.
Innovation Solution
An image processing apparatus that performs correspondence-driven object appearance transfer by aligning content features from the content image with style features from the style image to obtain a sparse correspondence map, allowing for the generation of a hybrid image that incorporates appearance and texture attributes from the style image while maintaining spatial structures from the content image.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If conventional image processing systems are used for appearance transfer, then the process is simple, but the system fails to capture detailed textures from the style image while retaining spatial structure from the content image when objects have complex geometries and extensive geometry variations
Solution Approach 1:
The patent introduces a correspondence map as an intermediary structure that establishes pixel-level relationships between content and style images. This correspondence map serves as a mediator that guides the texture transfer process, ensuring that detailed textures are accurately mapped from the style image to the content image while preserving the spatial structure, even when objects have complex geometries and extensive variations.
2Adaptability or versatility
If appearance transfer is performed on objects with complex geometries and extensive geometry variations, then the application scope is expanded, but the accuracy of texture capture and spatial structure retention deteriorates
Solution Approach 1:
The patent employs a dynamic approach by computing a correspondence map that adapts to the specific geometric variations between content and style images. Rather than using a fixed transformation, the system dynamically establishes pixel correspondences based on the actual geometric relationships in each image pair, allowing accurate texture transfer even when objects have complex geometries and extensive variations.
3Productivity
If a sparse correspondence map is used instead of dense correspondence, then the computational complexity is reduced, but the precision of texture alignment may be compromised
Solution Approach 1:
The patent applies partial action by using a sparse correspondence map that selects only the most critical pixel correspondences rather than computing all possible correspondences. This partial approach focuses computational resources on key alignment points that are sufficient for accurate texture transfer, achieving a balance between processing efficiency and alignment precision.
Data Source
AI summary
Systems and methods for image processing are configured. Embodiments of the present disclosure encode a content image and a style image using a machine learning model to obtain content features and style features, wherein the content image includes a first object having a first appearance attribute and the style image includes a second object having a second appearance attribute; align the content features and the style features to obtain a sparse correspondence map that indicates a correspondence between a sparse set of pixels of the content image and corresponding pixels of the style image; and generate a hybrid image based on the sparse correspondence map, wherein the hybrid image depicts the first object having the second appearance attribute.