Palmprint Image Generation Using Noise-Guided Multi-Scale Synthesis
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
The low generation efficiency of palmprint pictures due to the scarcity of multi-modal palmprint samples for training, resulting in poor recognition ability of palmprint picture matching models, is addressed by generating simulated palmprint pictures with noise addition and downsampling/upsampling operations.
Innovation Solution
A method and apparatus that generate palmprint pictures by combining simulated palmprint curves with noise vectors, performing downsampling and upsampling operations to create diverse palmprint pictures, enhancing the training efficiency of palmprint picture matching models.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If a large number of multi-modal palmprint sample pictures are used for training, then the recognition ability of the palmprint picture matching model is improved, but the training process becomes extremely dependent on the scale and diversity of sample pictures which are difficult to obtain due to privacy constraints
Solution Approach 1:
The patent uses a generator network to synthesize artificial palmprint pictures that replicate the characteristics of real palmprint images. The generator creates virtual sample pictures with diverse modalities (infrared, visible light, thermal) without requiring actual physical samples, thus solving the scarcity problem while maintaining training quality for improving recognition ability.
Solution Approach 2:
The patent employs an encoder-decoder architecture with noise vectors to transform input pictures into diverse output modalities. By manipulating latent space parameters and adding controlled noise during the encoding-decoding process, the system generates varied palmprint representations from a single input, enabling training with limited original samples.
2Productivity
If traditional training methods are used with limited samples, then the training process is simpler, but the generation efficiency of palmprint pictures is low and the scale and diversity of training samples are insufficient
Solution Approach 1:
The patent pre-trains an encoder-decoder framework before using it for picture generation. The encoder is first trained to extract features from real palmprint pictures, and then the decoder is trained to reconstruct images from these features. This preliminary training establishes a solid foundation that enables efficient generation of diverse samples without requiring complex retraining for each new sample set.
Solution Approach 2:
The patent introduces a latent space representation as an intermediary between the encoder and decoder. The encoder maps input pictures to latent vectors, and the decoder maps these latent vectors back to image space. This intermediary representation allows for efficient manipulation and generation of diverse palmprint samples by simply modifying the latent space inputs rather than requiring complex image processing pipelines.
Data Source
AI summary
A palmprint picture generation method including obtaining a simulated palmprint picture including a simulated palmprint curve, inputting the simulated palmprint picture and a preset first noise vector into a target palmprint picture generator, and performing a plurality of downsampling operations and a plurality of upsampling operations on the simulated palmprint picture in sequence through the target palmprint picture generator to generate a target palmprint picture.