Name-Face Matching Using Word Embedding Vectors
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current facial recognition systems rely on known face images for identification and have limited accuracy in matching faces with names without prior knowledge, with existing methods achieving only 59% accuracy using deep convolutional neural networks.
Innovation Solution
Implementing a system that maps names to vector spaces using word embedding techniques, allowing for the comparison of name vectors with face vectors to improve matching accuracy, achieving 72% to 80.5% accuracy through the application of various matching functions.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Adaptability or versatility
If deep convolutional neural networks are used to match faces with names, then the system can operate without prior knowledge of the person, but the accuracy is limited to 59%
Solution Approach 1:
The patent combines deep convolutional neural networks with word embedding techniques and matching functions (such as affine maps, logistic regression, neural networks, and recurrent neural networks) to create a hybrid system. This merging of multiple computational approaches allows the system to achieve 72-80.5% accuracy by leveraging the strengths of each component: face vector extraction from images, name vector representation through word embedding, and sophisticated matching functions to compare and rank name-face pairs.
2Reliability
If traditional facial recognition systems are used, then known face images can be stored and compared, but the system cannot identify unknown persons
Solution Approach 1:
Instead of storing known face images and comparing unknown faces against this database (traditional approach), the system inverts the problem by extracting face vectors from unknown face images and comparing them against name vectors derived from textual name data. This inversion enables the system to identify unknown persons by matching their facial features to associated names without requiring pre-stored reference images of the unknown individuals.
3Measurement precision
If word embedding is applied to map names to vector spaces, then matching accuracy improves to 72-80.5%, but the system complexity increases
Solution Approach 1:
The patent introduces name vectors as an intermediary representation that bridges the gap between textual name data and facial image data. By converting names into vector representations through word embedding techniques, the system creates a common mathematical space where names and face vectors can be meaningfully compared using matching functions. This intermediary vector representation simplifies the comparison process while enabling high-accuracy matching.
Data Source
AI summary
Described are methods, systems, and computer-program product embodiments for selecting a face image based on a name. In some embodiments, a method includes receiving the name. Based on the name, a name vector is selected from a plurality of name vectors in a dataset that maps a plurality of names to a plurality of corresponding name vectors in a vector space, where each name vector includes representations associated with a plurality of words associated with each name. A plurality of face vectors corresponding to a plurality of face images is received. A face vector is selected from the plurality of face vectors based on a plurality of similarity scores calculated for the plurality of corresponding face vectors, where for each name vector, a similarity score is calculated based on the name vector and each face vector. The face image is output based on the selected face vector.