Image Style Transformation via Object-Environment Coordination
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing image style transformation methods cannot effectively isolate and transform the style of a target object within an image, leading to monotonous and inharmonious results where the object's style does not coordinate with its surroundings.
Innovation Solution
An image generation method that extracts content and style features from both images, determines a target style feature by fusing overall and object style features, and generates a new image with content corresponding to the original content and style corresponding to the target feature, ensuring the object's style coordinates with its environment.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Adaptability or versatility
If overall image style transformation is applied, then the overall image style is transformed, but the target object style cannot be reflected independently
Solution Approach 1:
The patent segments the image processing into two distinct parts: overall image style transformation and target object style transformation. By dividing the style transformation task into global and local components, the system can independently control and preserve object style information while transforming the overall image style, thus resolving the contradiction between comprehensive style transformation and object style preservation.
Solution Approach 2:
The patent applies different style transformation qualities to different regions of the image. The target object region receives a different style treatment compared to the background regions. This local quality approach allows the object to maintain its distinctive style characteristics while the overall image achieves the desired style transformation, preventing loss of object style information.
2Adaptability or versatility
If overall image style transformation is applied, then the overall style changes, but the transformed image becomes monotonous and inharmonious
Solution Approach 1:
By segmenting the image into target object and background regions, the patent enables independent style control for each segment. This allows the system to create harmonious transitions between different style regions, avoiding the monotony that results from uniform overall style transformation. The segmented approach ensures that style changes are coordinated and aesthetically pleasing.
Solution Approach 2:
The patent introduces asymmetry in style transformation by applying different style parameters to the target object versus the background. This asymmetric treatment creates visual interest and harmony by allowing the object to stand out while maintaining coordination with the overall image style, preventing the inharmonious appearance that results from symmetric uniform transformation.
Data Source
AI summary
An image generation method includes: extracting a content feature of a first image; respectively extracting an overall image style feature of a second image, and an object style feature of a partial image block, which includes an object, in the second image; determining a target style feature at least according to the overall image style feature and the object style feature; and generating a third image according to the content feature and the target style feature.