Face Replacement Database Creation Using 3D Vector Angle Estimation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional face replacement techniques are cumbersome and time-consuming, requiring manual marking of feature points, are limited to static images, and often result in unnatural visual perception due to color differences and complex angle estimation processes.
Innovation Solution
A method for creating a face replacement database by estimating a 3D vector angle from a 2D face image, using detected corners of the eyes and mouth to form a quadrilateral, converting vertices to 3D coordinates, and matching these with a feature vector model to determine the rotation angle, allowing for automatic and precise dynamic image replacement with edge feathering for natural color adjustment.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If manual marking of feature points is used to calculate replacement regions and angles, then the replacement accuracy for static images can be achieved, but the operation becomes time-consuming and inconvenient for users
Solution Approach 1:
The system automatically detects feature points (corners of eyes and mouth) and calculates replacement parameters without user intervention. The computer vision algorithm processes the input image to identify key facial landmarks and computes the replacement region and rotation angle automatically, eliminating the need for manual marking while maintaining high precision
Solution Approach 2:
The manual mechanical process of marking feature points is replaced by an automated computer vision system that uses image processing algorithms to detect facial features and calculate geometric parameters, significantly reducing operation time while preserving measurement accuracy
2Adaptability or versatility
If conventional face replacement technique is applied, then single static image replacement can be achieved, but it is difficult to apply for dynamic images
Solution Approach 1:
The system is designed to handle both static and dynamic images by processing each frame independently. For video inputs, the algorithm extracts feature points and calculates replacement parameters for each frame, enabling dynamic image replacement while maintaining consistency with the original conventional method's accuracy for static images
3Productivity
If simple replacement method is used, then processing speed can be improved, but color difference occurs at the boundary resulting in unnatural visual perception
Solution Approach 1:
The system applies different processing strategies to different regions: the central replacement region uses direct image substitution for speed, while the boundary regions apply color matching and blending techniques to ensure natural transitions. This localized quality adjustment eliminates color differences at boundaries while maintaining overall processing efficiency
4Measurement precision
If complicated face angle estimation method is used, then estimation accuracy can be improved, but computation time increases significantly
Solution Approach 1:
The system extracts only the essential feature points (corners of eyes and mouth) needed for angle estimation, rather than using complex comprehensive facial landmark detection. By selecting and using only the critical subset of features, the system achieves sufficient estimation accuracy with significantly reduced computation time
Data Source
AI summary
A method for creating a face replacement database includes steps of creating a face database for storing a plurality of replaced images with a face image rotation angle by using a method for estimating a 3D vector angle from a 2D face image, and defining a region to be replaced in the replaced image. The method for estimating a 3D vector angle from a 2D face image includes the steps of creating a feature vector template; detecting a corner of eye and mouth in a face image; defining a sharp point in a vertical direction of the quadrilateral plane, and converting the vertices into 3D coordinates; computing the four vectors from the sharp point to the four vertices to obtain a vector set, and matching the vector set with the feature vector model to obtain an angle which is defined as a rotation angle of the input face image.