Tensor Processing Hardware for Fingerprint Data Acceleration
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing fingerprint processing technologies are inefficient in accelerating the generation of fingerprint representations and matching multiple fingerprints, particularly due to the lack of suitable hardware acceleration and parallelized execution capabilities.
Innovation Solution
Utilization of tensor processing hardware to accelerate fingerprint data processing by transforming 2D image patches and filters into pixel vectors and filter vectors, respectively, and performing matrix multiplication to implement convolutions, along with partitioning matching operations among multiple processing units and modulating clock frequencies to enhance processing efficiency.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Speed
If traditional software-based fingerprint processing is used, then implementation simplicity is maintained, but processing speed and computational efficiency are insufficient
Solution Approach 1:
The patent replaces traditional software-based processing with hardware acceleration using Tensor Processing Units (TPUs) and Graphical Processing Units (GPUs). The system transforms fingerprint images into tensor formats compatible with these hardware accelerators, enabling parallel computation of convolution operations that are prohibitively slow on conventional CPUs. This substitution of computational platform directly addresses the speed deficiency while accepting increased hardware complexity as a trade-off.
Solution Approach 2:
The patent converts two-dimensional fingerprint images into multi-dimensional tensor structures that can be processed by specialized hardware. By reshaping the data from 2D image format to 4D tensor format (batch, channels, height, width), the system enables parallel processing across multiple dimensions simultaneously, dramatically increasing computational throughput for fingerprint feature extraction and matching operations.
2Productivity
If sequential fingerprint matching is used, then processing simplicity is maintained, but time required for matching multiple fingerprints increases
Solution Approach 1:
The patent divides the fingerprint matching task into independent parallel units, where each processing unit handles a specific subset of fingerprints or specific convolution operations. The system segments the matching process by distributing different fingerprint templates across multiple processing units that can operate simultaneously, reducing the total time required to match multiple fingerprints while maintaining the complexity of coordinating these parallel operations.
Solution Approach 2:
The patent merges multiple sequential matching operations into a single parallel processing framework. By combining multiple convolution operations and fingerprint matching comparisons into unified tensor processing graphs that execute simultaneously on hardware accelerators, the system achieves high throughput for matching multiple fingerprints without requiring complex sequential coordination logic.
3Duration of action of moving object
If fixed clock frequency is used for processing, then system stability is maintained, but inability to meet maximum time requirements for matching operations
Solution Approach 1:
The patent implements dynamic clock frequency adjustment for the processing units based on the number of fingerprints to be matched and the current processing stage. The system transitions from fixed frequency operation to variable frequency operation, allowing the clock speed to be increased when fewer fingerprints need processing and decreased when more fingerprints require matching, thereby meeting maximum time requirements while optimizing power consumption. This dynamic adjustment introduces complexity in frequency control logic but enables flexible timing adaptation.
Data Source
AI summary
A method includes identifying, within an image that represents a fingerprint, a plurality of keypoints. The method also includes determining a plurality of pixel vectors by transforming, for each respective keypoint, a corresponding two-dimensional (2D) array of pixel values associated with the respective keypoint into a corresponding pixel vector, and obtaining a plurality of filter vectors that represent, for each respective filter of a plurality of filters, a transformation of a corresponding 2D array of filter values of the respective filter into a corresponding filter vector. The method additionally includes convolving the plurality of keypoints with the plurality of filters by multiplying, using a tensor processing hardware, a pixel matrix formed by the plurality of pixel vectors and a filter matrix formed by the plurality of filter vectors, and determining, based on the convolving, a representation of the fingerprint.