Recognition method and device for repairing abnormal vein images with generative adversarial network based on classification loss
A vein image and recognition method technology, which is applied in the fields of biometric feature recognition, biological neural network model, character and pattern recognition, etc., can solve the problem of poor repair effect of vein image, blurred vein texture structure, poor recognition performance of abnormal vein image, etc. question
- Summary
- Abstract
- Description
- Claims
- Application Information
AI Technical Summary
Problems solved by technology
Method used
Image
Examples
Embodiment 1
[0079]Referring to FIG. 1, a method of generating a classification loss to generate a network repairing abnormal vein image comprising the following steps:
[0080]1) Normal intravenous image B of several defect venous images and normal intravenous image B corresponding to a plurality of defect venous images, forming training sets. In this embodiment, since there is almost no suitable label image in reality, it is not possible to use a realistic abnormal venous image construction training set. Therefore, 200,000 Zhang Zhengyuan venous image B is selected, and it is multiplied with any mask image M (0. of the damaged portion of 0, the unwage portion is 1), resulting in 200,000 to break the intravenous image A, the intravenous image normalized size is 160 * 64 and here as a training set.
[0081]2) Construction of SK-RESNET-based generator networks, add SK-RESNET structures in the encoder of the generator, and the SK-RESNET structure is an embedded module of the SK module in the RESNET netw...
Embodiment 2
[0118]Referring to FIG. 6, the present embodiment relates to an identification apparatus for generating a network repairing abnormal vein image based on classification loss, including:
[0119]1) Normalization processing module for normalizing the normal intravenous image B of several defect venous images A and the normal vein image B corresponding to a plurality of defect venous images, forming training sets, normalized processing modules for implementing embodiments 1 Step 1) The function;
[0120]2) Generator network construction module for building SK-RESNET-based generator networks, add SK-RESNET structures in the encoder of the generator, and the SK-RESNET structure is an embedded SKNET module in the ResNet residual module. Module, generator network build module for implementing the function of step 2) of Example 1;
[0121]3) The discriminator network construction module is used to build a CNN-based discriminator network, and the discriminator network construction module is used to im...
experiment example
[0133]In order to verify the repair effect of the present invention to the abnormal vein image, 50 users' finger images are selected, each user 10 finger images, including a normal image library and a damaged image library (mirror dirty block) , A total of 50 * 10 * 2 = 1000 pairs, the image size is 500 * 200, and the image normalization is 160 * 64.
[0134]The damaged image of the damaged image library is separately repaired by the method generator repair module of the present invention and the conventional crinimisi method, and Fig. 7 A) means an intravenous image of the true dirty block, and Fig. 7B) is a modified finger vein image after the present invention. Fig. 7C) The traditional Crinimisi repair method refers to the intravenous image, which can obviously see the traditional crinimisi method has repaired the dirty block, but there are still some block effects, and the intravenous image grayscale has a fault, discontinuous However, the similarity of the same normal image simila...
PUM
Login to View More Abstract
Description
Claims
Application Information
Login to View More