Face Image Grouping via Non-Negative Matrix Factorization
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing methods for automatically creating image-based products face challenges in accurately grouping and processing large numbers of face images, particularly in identifying and prioritizing relevant individuals over strangers, which is time-consuming and inefficient.
Innovation Solution
A computer-implemented method that uses probabilistic criteria and non-negative matrix factorization to automatically categorize and group face images without prior training, allowing for scalable and flexible face analysis compatible with various face descriptors and similarity functions, effectively distinguishing between relevant and non-relevant faces.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Adaptability or versatility
If conventional face detection and grouping methods are used, then face images can be categorized into groups, but the methods require prior training with known persons' faces and are not scalable to large user accounts with hundreds of thousands to millions of photos
Solution Approach 1:
The system performs self-service by automatically grouping face images without requiring prior training with known persons' faces. The method uses probabilistic criteria and non-negative matrix factorization to autonomously categorize faces into groups based on similarity metrics, eliminating the need for manual supervision or pre-training datasets.
Solution Approach 2:
The invention changes the fundamental parameters of face grouping by transitioning from supervised learning approaches to unsupervised probabilistic methods. Specifically, it employs non-negative matrix factorization to transform the face image data into latent factors that capture similarity relationships, allowing scalable grouping without traditional training parameters.
2Measurement precision
If all face images in a large user account are processed together, then comprehensive grouping can be achieved, but the computation time and processing resources become prohibitively large
Solution Approach 1:
The system segments the large set of face images into smaller batches or chunks for processing. By dividing the comprehensive dataset into manageable segments, the method maintains grouping accuracy while reducing the computational burden and time required for processing each individual batch.
Solution Approach 2:
The method applies partial action by processing face images in multiple passes or iterations rather than attempting to process all images simultaneously. This approach achieves comprehensive grouping accuracy through repeated partial processing of different subsets of the data.
3Measurement precision
If traditional face recognition methods are used, then faces can be identified, but the methods assume single-face regions are already detected and extracted, requiring additional preprocessing steps
Solution Approach 1:
The system achieves universality by creating a face grouping method that works directly with detected face regions without requiring specific preprocessing assumptions. The probabilistic approach with non-negative matrix factorization can handle various input formats and face detection methods, making the solution universally applicable across different detection systems.
Data Source
AI summary
A computer-implemented method of grouping faces in large user account for creating an image product includes adding the face images obtained from an image album in a user's account into a first chunk; if the chunk size of the first chuck is smaller than a maximum chuck value, keeping the face images from the image album into the first chunk; otherwise, automatically separating the face images from the image album into a first portion and one or more second portions; keeping the first portion in the first chunk; automatically moving the second portions to subsequent chunks; automatically grouping face images in the first chunk to form face groups; assigning the face groups to known face models associated with the user account; and creating a design for an image-based product based on the face images in the first chunk associated with the face models.