Face Recognition Using Sparse Binary Vectors for Sub-Linear Search
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing face recognition systems are inefficient and unreliable in processing large volumes of image data due to the use of human-designed algorithms that create large and slow-to-compare representations of faces, making it difficult to perform accurate searches across vast databases.
Innovation Solution
The development of systems that utilize deep neural networks (DNNs) to generate sparse feature vectors, which are then binarized into compact binary vectors, enabling sub-linear search methods for efficient comparison against large databases, reducing the need for exhaustive linear searches.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If human-designed algorithms are used to create face representations, then the representations are reliable in differentiating faces, but the representations are large and slow to compare
Solution Approach 1:
The patent transforms face representations from continuous floating-point vectors to discrete binary vectors through quantization. This parameter change reduces the data size and enables faster comparison operations while maintaining sufficient discrimination capability for face recognition tasks.
Solution Approach 2:
The patent extracts only the essential discriminative features needed for face recognition and represents them in a compact binary format. By taking out only the critical information and discarding redundant data, the system achieves faster search speeds while preserving reliability.
2Reliability
If large databases of face images are searched using traditional methods, then comprehensive search coverage is achieved, but exhaustive linear search requires excessive time
Solution Approach 1:
The patent segments the large database of face images into smaller clusters or groups based on binary representation similarities. This segmentation allows the search algorithm to navigate through the database more efficiently by first identifying relevant segments and then performing detailed searches only within those segments, rather than conducting exhaustive linear search across the entire database.
Solution Approach 2:
The patent performs preliminary binarization and indexing of face representations before the actual search operation. By pre-processing the data into a compact binary format and organizing it appropriately, the system prepares the database for rapid querying, significantly reducing the time required for comprehensive searches.
3Productivity
If compact binary vectors are used for face representation, then sub-linear search becomes feasible, but the representation may lose some accuracy
Solution Approach 1:
The patent carefully designs the binarization process to transform continuous features into discrete binary values while preserving the most discriminative information. By optimizing the parameter transformation process, the system achieves compact representation that enables sub-linear search without excessive loss of recognition accuracy.
Solution Approach 2:
The patent creates binary copies of face representations that serve as efficient proxies for the original continuous vectors. These binary copies enable rapid search operations while the system maintains the ability to refer back to more precise representations when needed for final verification.
Data Source
AI summary
Systems and methods are provided for performing automated face recognition and comparison. An input image of a face may be received and cropped, and the image may be processed through a deep neural network (DNN) to produce a k-dimensional feature vector. The k-dimensional feature vector may be converted to a k-dimensional binary vector by transforming each value in the vector to either 1 or 0. To search for nearest matches of the image in a database of gallery images of faces, the system may compare sub-strings of the binary vector to hash tables created from sub-strings of the gallery images, enabling sub-linear searching that allows locating the closest matches from among the entire gallery without requiring an exhaustive linear search of the entire gallery.