CN-121998890-A - Medical fingerprint image quality automatic evaluation method based on deep learning

Abstract

The fingerprint image quality automatic evaluation and enhancement method based on deep learning realizes high-efficiency and accurate quality evaluation and classification, provides a self-adaptive image enhancement function, improves the usability of low-quality fingerprint images, and ensures the accuracy and reliability of medical feature extraction.

Inventors

• Lv Shixuan
• YUAN JIAMIN

Assignees

• 丰格菩瑞慈科技(无锡)有限公司

Dates

Publication Date

20260508

Application Date

20241106

Claims (1)

1. 1. The fingerprint image quality automatic evaluation and enhancement method based on deep learning realizes high-efficiency and accurate quality evaluation and classification, provides a self-adaptive image enhancement function, improves the usability of low-quality fingerprint images, and ensures the accuracy and reliability of medical feature extraction.

Description

Medical fingerprint image quality automatic evaluation method based on deep learning Technical Field The invention belongs to the technical field of image processing and medical image analysis, and particularly relates to an automatic fingerprint image quality assessment method based on deep learning, which is applied to the industries of medical health monitoring, disease screening, personalized medical management and the like Background In medical applications, fingerprint images are widely used as a non-invasive, personalized biometric feature for health monitoring, disease screening, and personalized medical management. However, the quality of fingerprint images is often affected by a number of factors, such as the quality of the acquisition device, the ambient light at the time of acquisition, the condition of the finger surface (e.g., humidity, stains), the acquisition angle, etc. These factors can cause problems such as noise, blurring, strong light reflection, broken lines and the like in the fingerprint image, and directly influence the extraction and analysis of subsequent medical features. Disclosure of Invention 1. System architecture and working principle The workflow of the system relates to the following main modules of an image input and preprocessing module, a deep learning quality evaluation module, an image enhancement processing module and a reevaluation and output module 2. Image input and preprocessing module Step 1 (image input) the system receives a fingerprint image, which may be a gray scale or color image, acquired by the fingerprint acquisition device. Because medical applications require high image quality, the images may contain noise, blurring, etc. problems that affect subsequent processing. Thus, the image is first preprocessed after input. Step2 (pretreatment) the pretreatment stage comprises the following steps: And (3) graying, namely converting the color image into a gray image, reducing the calculation complexity, and simultaneously retaining key information such as fingerprint lines and the like. And (3) standardization, namely carrying out standardization processing on the size and the pixel value of the input image, and ensuring consistency of input data in the subsequent model processing process. Through the preprocessing steps, the system can process fingerprint images acquired under different resolution and different illumination conditions. 3. Deep learning quality evaluation module Step 3 (quality assessment model training) the invention builds a deep learning quality assessment model based on Convolutional Neural Network (CNN). The data sets required for model training fall into two categories: High-quality image, clear, noiseless, blurry and uniform illumination. Low quality images including noise, blur, uneven illumination, low contrast, and the like. Recording image paths using CSV files and labels therefor: image normalization-the image is resized to a fixed size (e.g., 224x 224) and the pixel values are normalized. And data enhancement, namely rotating, zooming, turning and the like are carried out on the image, so that more quality problems are simulated, and the generalization capability of the model is enhanced. And (3) performing image quality assessment by using a Convolutional Neural Network (CNN), wherein the model consists of a plurality of convolutional layers, a pooling layer and a full-connection layer, and extracting the characteristics of the image layer by layer. Dividing the data set into a training set and a verification set, and dividing according to the proportion of 80% training set and 20% verification set Model training Training the model by using the training set, iteratively adjusting model parameters, and optimizing the model by using a back propagation algorithm. ( Loss Function (Loss Function) using a cross entropy Loss Function (categorical crossentropy) is applicable to classification problems. And an Optimizer (Optimizer) is used for automatically adjusting the learning rate by using an Adam Optimizer and improving the training efficiency. ) Model evaluation After training is completed, the accuracy of the model is assessed using the validation set, checking its classification ability for high quality and low quality images. Step4 (image quality assessment) after training is completed, the system performs quality assessment on the input fingerprint image. 4. Image enhancement processing module Step 5 (low quality image enhancement) for the image assessed as low quality, the system automatically applies an image enhancement algorithm aimed at improving the sharpness and detail of the image. The specific enhancement steps comprise: Deblurring, namely enhancing the line details of the blurred image through a deconvolution (deconvolution) algorithm. First, a blur kernel (Point Spread Function, PSF) is estimated, which is deconvoluted from the blurred image using a deconvolution algorithm (Wiener deconvolution), restoring the detail