Face Image Defect Removal Preserving Skin Texture
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current image processing methods, such as deep learning algorithms, often result in unnatural or one-sided processed images by globally changing pixels, leading to a lack of skin texture and quality in face images, especially when trying to remove defects like pimples or moles.
Innovation Solution
An image processing method and apparatus that uses a face detection algorithm to identify defect elements in a face image and applies an image processing model trained with samples having a face distortion degree below a threshold, allowing for targeted removal of defects while preserving skin texture, using a generative adversarial network to convert the image into a target face image without the defect.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Manufacturing precision
If deep learning algorithms are used to globally change pixels to remove defects, then defect removal capability is improved, but skin texture and natural appearance deteriorate
Solution Approach 1:
The patent segments the face image into multiple regions based on defect location and characteristics. Instead of globally processing the entire image, the system identifies specific defect regions (pimples, moles, etc.) and processes only those localized areas while preserving the rest of the skin texture and facial features. This regional segmentation approach enables precise defect removal without compromising overall skin appearance.
Solution Approach 2:
The patent applies different processing qualities to different regions of the image. Defect regions receive intensive processing to remove unwanted elements, while non-defect skin regions maintain their original texture and characteristics. This local quality differentiation ensures that defect removal is achieved without creating unnatural uniformity across the entire face.
2Productivity
If global pixel modification is applied to remove defects, then processing speed is improved, but image quality and natural characteristics deteriorate
Solution Approach 1:
The patent extracts only the necessary processing operations from the entire image. Instead of applying transformations to all pixels globally, the system extracts and processes only the pixel regions containing defects. This extraction approach maintains processing efficiency while significantly improving image quality by avoiding unnecessary modifications to healthy skin areas.
Solution Approach 2:
The patent applies partial action by processing only the portions of the image that contain defects rather than the entire image. This selective processing reduces computational overhead while achieving the desired defect removal effect, thereby maintaining productivity without sacrificing image quality.
3Productivity
If training samples with high face distortion are used, then model training speed is improved, but processing accuracy and natural appearance deteriorate
Solution Approach 1:
The patent changes the parameter of face distortion degree in training sample selection. By setting a threshold on the face distortion degree parameter, the system selects training samples that balance processing accuracy requirements with training efficiency. This parameter-based filtering ensures that only appropriately distorted faces are used for training, avoiding excessive distortion that would compromise model accuracy.
Data Source
AI summary
An image processing method, performed by a computer device, includes: obtaining an input image to be processed; performing face detection on the input image to obtain an input face image to be processed including at least one defect element relating to a skin element; and inputting the input face image into an image processing model to obtain a target face image corresponding to the input face image without a first defect element amongst the at least one defect element, wherein a training sample of the image processing model includes a first face image with a first face distortion degree less than a preset threshold and that is annotated with the first defect element.