Style Propagation via Parameter Copying for Mobile Devices
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional content editing systems fail to effectively propagate styles across digital images, especially when dealing with different scenes, and require extensive training data and computational resources, making them unsuitable for mobile devices.
Innovation Solution
A learned style system that generates style data from a limited set of modified images, using image representations and modification parameters to apply styles to input images, allowing for efficient style propagation without the need for extensive training data or resources.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Manufacturing precision
If conventional systems use machine learning with trained neural networks to propagate styles, then the style propagation accuracy is improved, but the training time and computing resources required increase significantly
Solution Approach 1:
The patent copies style parameters directly from reference images with similar content characteristics instead of using complex neural networks. It identifies images with similar content and copies their style parameters (exposure, contrast, saturation, etc.) directly, achieving efficient style propagation without extensive training time.
Solution Approach 2:
The patent changes the approach from learning complex transformations to directly adjusting visual parameters. It modifies style parameters (exposure, contrast, highlights, shadows, saturation, etc.) based on content similarity metrics, achieving accurate style propagation through parameter adjustment rather than neural network transformation.
2Manufacturing precision
If conventional systems use trained neural networks for style propagation, then the visual quality of styled images is improved, but the computing resources required become prohibitive for mobile devices
Solution Approach 1:
The patent replaces expensive, resource-intensive neural networks with simple, lightweight parameter copying operations. It uses basic image comparison metrics and direct parameter assignment, which consume minimal computing resources and can run efficiently on mobile devices without requiring powerful hardware.
Solution Approach 2:
The patent extracts only the essential style parameters (exposure, contrast, saturation, etc.) from reference images and applies them directly, rather than using entire neural network models. This extraction approach maintains visual quality while dramatically reducing computing resource requirements.
3Productivity
If simple parameter copying is used to propagate styles, then the computational efficiency is improved, but the ability to handle different scenes (e.g., sky vs. indoor) deteriorates
Solution Approach 1:
The patent applies different style parameters selectively based on content type. It identifies specific content characteristics (sky, indoor scenes, portraits, etc.) and applies appropriate style parameters from matching reference images, ensuring that each scene receives locally optimized style treatment rather than uniform parameter copying.
Solution Approach 2:
The patent segments the style propagation process into content-based categories. It divides reference images and target images into content groups (sky scenes, indoor scenes, landscapes, etc.) and performs parameter copying within matching segments, improving both efficiency and scene adaptability through structured organization.
4Manufacturing precision
If extensive training data is used to train neural networks for style propagation, then the learning accuracy is improved, but the data requirements and training complexity increase
Solution Approach 1:
The patent uses partial action by selecting only the most relevant reference images based on content similarity rather than using extensive training datasets. It identifies and uses a small subset of highly relevant reference images for parameter copying, achieving accurate style propagation without the need for large volumes of training data.
Data Source
AI summary
Image modification styles learned from a limited set of modified images are described. A learned style system receives a selection of one or more modified images serving as a basis for a modification style. For each modified image, this system creates a modification memory, which includes a representation of the image content and modification parameters describing modification of this content to produce the modified image. These modification memories are packaged into style data, used to apply the modification style to input images. When applying a style, the system generates an image representation of an input image and determines measures of similarity between the input image's representation and representations of each modification memory in the style data. The system determines parameters for applying the modification style based, in part, on these similarity measures. The system modifies the input image according to the determined parameters to produce a styled image.