Hairstyle Transformation via Grid Deformation

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Existing methods for transforming hairstyles, such as affine transformation and large-scale modification of the CycleGAN training data set, often result in distorted outputs and are labor-intensive, unstable, and inefficient, especially when dealing with complex hairstyles.

Innovation Solution

The method determines a face bounding box based on face key points, constructs grids within this box, and deforms only the edge lines of the hairstyle using a target hairstyle function to obtain a deformed grid curve, avoiding the need to transform the entire face image and thus preventing distortion.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Ease of operation

If affine transformation is used to deform the hairstyle, then the hairstyle can be transformed, but the output becomes distorted

Engineering Contradiction:
Improvehairstyle transformationVSAvoidoutput accuracy
Core Design Contradiction:
Ease of operationVSManufacturing precision

Solution Approach 1:

The method segments the face image into multiple grid regions, with each grid independently deformed according to the hairstyle transformation requirements. This localized segmentation allows precise control over hairstyle deformation while preserving the integrity of the underlying face features, resolving the contradiction between ease of transformation and output accuracy.

Inventive Principle:
Principle #1Segmentation

2Ease of operation

If the entire face image is transformed to change hairstyle, then the hairstyle can be deformed, but the face image becomes distorted

Engineering Contradiction:
Improvehairstyle deformationVSAvoidface image distortion
Core Design Contradiction:
Ease of operationVSObject-affected harmful factors

Solution Approach 1:

The method extracts and isolates the hairstyle region from the rest of the face image by constructing grids that cover only the hair area. By applying deformation operations exclusively to these grid regions containing the hairstyle, the face image remains unaffected and undistorted, while achieving the desired hairstyle transformation.

Inventive Principle:
Principle #2Taking out (Extraction)

3Adaptability or versatility

If large-scale modification of CycleGAN training data set is performed, then hairstyle transformation can be achieved, but labor cost and time increase significantly

Engineering Contradiction:
Improvehairstyle transformation capabilityVSAvoiddata preparation time
Core Design Contradiction:
Adaptability or versatilityVSLoss of time

Solution Approach 1:

Instead of modifying large amounts of training data, the method creates a virtual deformation model by constructing grids over the hairstyle region and applying mathematical transformation functions directly to these grids. This copying approach generates transformation results without requiring extensive data preparation, significantly reducing labor cost and time while maintaining versatility in hairstyle transformation.

Inventive Principle:
Principle #26Copying

4Manufacturing precision

If manual modification of training data is performed to obtain accurate hairstyle deformation, then deformation accuracy improves, but productivity decreases

Engineering Contradiction:
Improvedeformation accuracyVSAvoidprocessing efficiency
Core Design Contradiction:
Manufacturing precisionVSProductivity

Solution Approach 1:

The method replaces manual mechanical modification of training data with an automated mathematical system. By using grid construction and transformation functions (such as thin-plate spline or radial basis function interpolation), the system automatically achieves accurate hairstyle deformation without manual intervention, thereby maintaining high precision while significantly improving processing efficiency and productivity.

Inventive Principle:
Principle #28Mechanics substitution (Replace mechanical system)

Data Source

PatentEP3855386B1Method, apparatus, device and storage medium for transforming hairstyle and computer program product
Publication Date: 2023.09.20 BEIJING BAIDU NETCOM SCI & TECH CO LTD
  • EP3855386B1 patent drawingFigure 1
  • EP3855386B1 patent drawingFigure 2~3
  • EP3855386B1 patent drawingFigure 4~5

AI summary

The present disclosure discloses a method, apparatus, device, and storage medium for transforming a hairstyle and a computer program product, and relates to the fields of image processing technology, augmented reality technology, and deep learning technology. A specific implementation includes: determining a face bounding box according to information on face key points of acquired face image; constructing grids according to the face bounding box; deforming, by using an acquired target hairstyle function, edge lines of at least a part of the constructed grids, which comprises the hairstyle, to obtain a deformed grid curve; determining a deformed hairstyle in the face image according to the deformed grid curve. The present disclosure employs a target hairstyle function to deform the edge lines of at least part of the grids, which comprises the hairstyle, and does not need to deform the face image as a whole and an output is accurate, and thus the deformed hairstyle in the face image is not distorted.