Generative Hair Pixel Compositing Across Edited Face Crop Boundaries

Resolve Bottlenecks,
Find Innovative SolutionsGenerate Solutions

Solution Overview

Problem

Conventional digital image systems struggle with accuracy in modifying digital images outside of an edited image crop, particularly in areas like hair pixels, leading to unrealistic and visually jarring results with clear breaks between edited and unedited regions.

Innovation Solution

A hair blending system utilizing a generative neural network, combined with instance segmentation, part segmentation, and deep matting models, generates replacement hair pixels that seamlessly integrate with unedited areas, ensuring smooth transitions across crop boundaries.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Productivity

If conventional digital image systems apply image modifications only within a face region crop, then the editing process is simple and fast, but the accuracy and realism of the modified image deteriorates due to clear breaks between edited and unedited regions

Engineering Contradiction:

Improveediting speedVSAvoidimage modification accuracy

Core Design Contradiction:

ProductivityVSManufacturing precision

Solution Approach 1:

The system segments the image into different regions (face region with modification and surrounding regions without modification) and applies different processing strategies to each. The generative neural network is specifically trained to generate hair pixels that bridge these segments, creating seamless transitions while maintaining the simplicity of the editing process.

Inventive Principle:

Principle #1Segmentation

Solution Approach 2:

The generative neural network acts as an intermediary between the edited face region and the unedited surrounding regions. It generates intermediate hair pixels that blend the modified face area with the unmodified areas, eliminating the clear breaks and creating a natural transition without requiring complex processing of the entire image.

Inventive Principle:

Principle #24Intermediary (Mediator)

2Device complexity

If conventional systems use basic image processing for hair modification, then the device complexity is low, but the visual quality and realism of the output deteriorates

Engineering Contradiction:

Improvesystem complexityVSAvoidvisual quality

Core Design Contradiction:

Device complexityVSReliability

Solution Approach 1:

The system changes the fundamental parameters of hair pixel generation by using a generative neural network instead of basic image processing. This allows the system to create realistic hair pixels with complex patterns, textures, and lighting effects that match the surrounding hair, significantly improving visual quality while the automated nature of the process keeps the system manageable.

Inventive Principle:

Principle #35Parameter changes

Solution Approach 2:

The patent replaces traditional mechanical image processing techniques with a generative neural network approach. This substitution enables the system to understand and replicate complex hair structures, lighting conditions, and textures that would be extremely difficult to achieve with conventional algorithms, thereby improving visual quality without proportionally increasing complexity.

Inventive Principle:

Principle #28Mechanics substitution (Replace mechanical system)

3Stability of the object's composition

If the system generates replacement hair pixels for entire hair regions outside the face crop, then the image continuity improves, but the computational time and processing resources increase

Engineering Contradiction:

Improveimage continuityVSAvoidprocessing time

Core Design Contradiction:

Stability of the object's compositionVSLoss of time

Solution Approach 1:

The system applies local quality by generating replacement hair pixels only for specific regions where continuity is needed, rather than processing the entire hair area. The generative neural network focuses computational resources on the boundary regions between edited and unedited areas, creating high-quality transitions only where necessary to maintain image continuity.

Inventive Principle:

Principle #3Local quality

Solution Approach 2:

The system uses partial action by generating replacement hair pixels only for the portions outside the face crop that require continuity improvement, rather than regenerating all hair pixels. This selective approach maintains image continuity in critical areas while avoiding unnecessary computational processing in regions where the original hair pixels are already sufficient.

Inventive Principle:

Principle #16Partial or excessive action

Data Source

PatentUS12475623B2Generating and compositing hair pixels using generative neural networks

Publication Date: 2025.11.18 ADOBE INC

AI summary

The present disclosure relates to systems, methods, and non-transitory computer readable media for generating and composting pixels of a digital image that depict hair of an individual using generative neural networks. In some embodiments, the disclosed systems receive a modification to a face crop enclosing a face depicted within a digital image. In some cases, the disclosed systems determine, from the modification, modified hair pixels within the face crop of the digital image and unmodified hair pixels outside of the face crop of the digital image. The disclosed systems generate, for the unmodified hair pixels outside of the face crop, replacement hair pixels that resemble the modified hair pixels utilizing a generative neural network. Additionally, the disclosed systems generate a modified digital image by replacing the unmodified hair pixels outside of the face crop with the replacement hair pixels.