Portrait Beautification via Texture Rendering and Skin Smoothing
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Traditional skin grinding and beautifying algorithms for selfies often result in a loss of facial texture, leading to an unnatural appearance, and sharpening operations introduce image noise, reducing the effectiveness of blemish removal and enhancing the sense of an unreal face.
Innovation Solution
A method that acquires a facial texture image by matching skin texture material with the face area of a target object, incorporating facial key point and posture information, and fuses this image with a smoothed skin region to enhance the natural appearance and realism of portrait beautification.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Stability of the object's composition
If high-intensity low-pass filtering is performed on the face to improve smoothness and uniformity, then skin grinding effect is achieved, but facial texture is lost resulting in unnatural appearance
Solution Approach 1:
The patent segments the face processing into two distinct stages: first applying low-pass filtering to smooth the skin region and remove blemishes, then separately generating a facial texture image through rendering skin texture material. This segmentation allows each stage to optimize for its specific function without interfering with the other, thus preserving texture while achieving smoothness.
Solution Approach 2:
The patent introduces an intermediary facial texture image generated by rendering skin texture material that matches the face area. This intermediary element serves as a bridge between the smoothed skin region and the original facial texture, allowing texture information to be restored without reintroducing blemishes or noise.
2Loss of information
If sharpening operations are performed to restore facial texture, then texture details are enhanced, but image noise is introduced reducing blemish removal effectiveness
Solution Approach 1:
Instead of using traditional sharpening operations that directly modify the smoothed image and introduce noise, the patent uses an intermediary rendering approach. Skin texture material is rendered to create a facial texture image that inherently contains realistic texture details without the harmful side effects of sharpening filters, thus avoiding noise introduction while restoring texture.
Solution Approach 2:
The patent creates a copy of facial texture information through rendering skin texture material rather than attempting to recover it from the smoothed image through sharpening. This copying approach generates realistic texture details from scratch based on the rendered material, avoiding the noise amplification that occurs when trying to enhance details from already-smoothed data.
3Stability of the object's composition
If traditional skin grinding is applied uniformly across the face, then overall smoothness is improved, but facial key point alignment and posture accuracy are compromised
Solution Approach 1:
The patent applies local quality by processing the skin region and facial texture separately with different operations. The skin region undergoes uniform low-pass filtering for smoothness, while the facial texture is generated through rendering that specifically matches facial key point information and posture information. This allows each region to have optimized properties without compromising the other.
Solution Approach 2:
The patent segments the facial processing into distinct components: skin region smoothing and facial texture generation. The facial texture image is specifically designed to match facial key point and posture information, ensuring precision in these areas while the skin region achieves overall smoothness through separate processing.
Data Source
AI summary
A method for processing images is provided. The method includes: acquiring a first image by smoothing a skin region of a target object in an original image; determining a skin texture material matching with a face area of the target object; acquiring a facial texture image of the target object by rendering the skin texture material; and acquiring a second image by fusing the facial texture image with the first image.