Face Image Style Transfer via Region-Specific Parameter Adjustment
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current image style transfer methods fail to accurately transform facial features such as hair, eyes, and nose from a real face image into a virtual character image, resulting in low-level simulation and overall color similarity rather than detailed feature transformation.
Innovation Solution
An image processing method that acquires a real face image and a virtual face image, extracts features, calculates similarity, adjusts target parameters based on similarity to generate a target face image with enhanced feature transformation, and renders the target face image with improved detail and style transfer.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If image style transfer model is used to transform overall style of content image, then overall color similarity is achieved, but detailed feature transformation (hair, eyes, nose shape) cannot be transformed to virtual character style
Solution Approach 1:
The patent segments the face image into multiple regions (skin, hair, eyes, nose, mouth) and applies different processing strategies to each region. By dividing the transformation task into region-specific sub-tasks, the system can achieve both overall color similarity and detailed feature transformation accuracy simultaneously.
Solution Approach 2:
The patent applies local quality by using region-specific color palettes and transformation parameters for different facial areas. Each region (e.g., hair, skin, eyes) has its own color characteristics that are adjusted independently to match the virtual character style while maintaining the overall face coherence.
2Manufacturing precision
If target parameters are adjusted to improve feature transformation accuracy, then detailed feature similarity increases, but computational complexity and processing time increase
Solution Approach 1:
The patent performs preliminary action by pre-defining region color palettes and transformation rules before the actual image processing. This preparation work includes identifying facial regions and establishing color relationships, which simplifies the subsequent parameter adjustment process and reduces computational complexity during runtime.
Solution Approach 2:
The patent uses parameter changes by adjusting color palette parameters (color values, saturation, brightness) and region parameters (mask weights, blending ratios) to control the transformation process. These parameter adjustments enable fine-tuned control over feature transformation accuracy while maintaining manageable system complexity.
3Manufacturing precision
If region color palettes are used to transform facial features, then detailed feature transformation accuracy improves, but processing time increases due to multiple region calculations
Solution Approach 1:
The patent applies partial action by focusing computational resources on key facial regions that have the greatest impact on visual similarity (such as hair, eyes, and skin). Not all regions are processed with the same level of detail, which optimizes the balance between transformation accuracy and processing efficiency.
Solution Approach 2:
The patent merges the processing of multiple facial regions into a unified transformation framework. By combining region-specific color palette adjustments with overall face style transfer, the system achieves detailed feature transformation while reducing redundant calculations and improving processing efficiency.
Data Source
AI summary
An image processing method and apparatus, a storage medium and an electronic device are disclosed. The method comprises: acquiring a first face image in a real scene and a second face image in a virtual scene, wherein the second face image is generated at least according to a first value of at least one target parameter, and the at least one target parameter is used for determining at least one attribute of a face image in the virtual scene; extracting at least one first image feature from the first face image, and extracting at least one second image feature from the second face image; acquiring a similarity between the at least one first image feature and the at least one second image feature; adjusting the first value of the at least one target parameter based on the similarity to obtain a target value of the at least one target parameter; and rendering the at least one target parameter according to the target value to obtain the target face image in the virtual scene.