Finger Vein Biometric Identification using Convolutional Neural Network

The term finger vein recognition refers to the process of authenticating the human’s identity by using biometric techniques. Deep learning methods are widely used in these areas to perform well. A convolutional neural network is one of the deep learning methods which is effectively used for finger vein biometric identification. This article is going to briefly enumerate the aspects of finger vein biometric identification using convolutional neural network. At the end of the article, you could extremely be benefited by us to be sure.

Generally, finger impressions are used to identify human beings. As the population is increasing it is difficult to identify each and every individual with their images. Thus finger vein identification exactly matches with the individual by a pre-registered database. The database consists of humans’ finger veins and finger impressions. In fact, every human being has unique finger veins. Our technical team has initiated this article with the key features of the finger vein biometric identification for ease of your understanding. Can we have them? Come let us try to understand them. 

Key Features of Finger Vein Biometric Identification

• Effective authentication process without missing any traces
• Active authentication process even in skin cracks or rashes
• Low finger vein manipulations in the individual’s physical condition
• Constant finger vein patterns don’t require frequent enrollments
• Least possibilities to steal/read the immense veins
• Fewer interferences of finger/hand impressions
• Zero false acceptance rate by finger vein uniqueness

The above listed are some of the unique features of finger vein biometric identification. In fact, finger veins cannot be duplicated in real-time even identical twin’s finger veins also vary. Finger veins never get dynamic because they are constant in nature thus enrolling finger impressions once is enough. As the matter of fact, veins are situated inside of the human body. So doing forgery in this area could benefit the unsuccessful attempts to the bad guys.

Generally, compared to other biometric techniques finger vein technologies is less invasive. We hope that you would have understood the things listed as of now. Vein recognition is the authentication system of individuals to recognize their unique identity. Vascular patterns are matched here for recognition. Let’s have further explanations about the recognition used for the ease of your understanding.

“This is the article which is fully contented with the concepts of finger vein biometric identification using convolutional neural network” 

What is Vein Recognition used for?

• Image Artifacts (Noises) Removals
• Image Smoothing & Normalization
• Image Restoration & Contrast Enrichments
• Image Resizing & Rotation
• Image Cropping & Realignments
• Image Color Conversions (RGB to Gray)

These are the tasks and reasons behind using finger vein recognition technologies. If you still need any clarifications in these areas feel free to approach our researchers. They are always delighted to help in the fields of technology. As we are conducting various researches and projects we do know each and every method formerly and currently used in the concepts.

Similar to other concepts, we are here going to highlight the traditional methods that are used for finger vein recognition for your better understanding in these areas. We are going to list out some of the important methods so paying your kind attention here will yield you the former perceptions in the finger vein recognition. Are you guys ready to know about that? Come let’s try to understand them with clear handy notes. 

Traditional Methods for Finger Vein Recognition 

• Vein Line Tracking
  • It considers the vein line features & ROI data textures are not used
  • Blood vessel features ensure the recognition accuracy
• Edge Preserving Elliptical High Pass Filters & Gabor Filters
  • It enriches the blood vessel direction’s images
  • Achieves accuracy by acquiring high-quality images
• Personalized Best Bit Map & Local Binary Pattern
  • These methods consider the blood vessels local pattern
  • It increases the processing speed by removing the vein line detection (ROI)

These are the traditional methods used in finger vein recognition. They are abundantly used for human identity recognition. Even though, all these methods are failed to perform well because of finger vein light shades and their misalignments. For this reason, top engineers in the world are using the convolutional neural network in finger vein biometric identification. We can overcome these limitations by applying deep learning  convolutional neural network method. In this regard, we would like to showcase the important issues that arise in finger vein biometric identification.  

Important Issues in Finger-vein Biometric Identification

• Lighting Intensity Imbalances
• Ineffective Pixel Translations
• Low Amount of Finger-vein Patterns

In proper cases, finger vein recognition is dealt with the single texture images. Besides, it has consisted of numerous artifacts and variations. Accuracy of the finger vein recognition is minimized by these issues. However, we can deploy the CNN methods to enhance the progressions.

In fact, our researchers are very familiar with these areas and you can have our suggestion in this area. In the following passage, we have listed how does CNN recognize the finger veins step by step with clear hints to make you understand.

How Does CNN Recognize the Finger Veins?

• Step1
  • Finger vein devices collect the finger impressions of the users
• Step2
  • Devices convert the impressions into images
• Step 3
  • Extracts the features in the forms of vectors
• Step 4
  • Estimates the Euclidean distances of the 2 vectors
• Step 5
  • Displays the results as authentication success/failure

These are the steps involved in CNN vein recognition. Here, we thought that showing the structure of the convolutional neural network will educate you on something new. Yes, we are going to cover the next section with the structure of CNN. Shall we get into the next phase? Come let us try to understand them.

Structure of CNN

• Input & Conv2D Layer Types
  • Conv2D_1
    • Input: (75,75,3)
    • Output: (73,73,64)
  • Conv2D_2
    • Input: (36,36,64)
    • Output: (34,34,64)
  • Conv2D_3
    • Input: (17,17,64)
    • Output: (15,15,64)
  • Conv2D_1_Input
    • Input: (75,75,3)
    • Output: (75,75,3)
• Max Pooling 2D Layer Types
  • MAX_Pooling2D_1
    • Input: (73,73,64)
    • Output: (36,36,64)
  • MAX_Pooling2D_2
    • Input: (34,34,64)
    • Output: (17,17,64)
  • MAX_Pooling2D_3
    • Input: (15,15,64)
    • Output: (7,7,64)
• Dense Layer Types
  • Dense_1
    • Input: (3136)
    • Output: (500)
  • Dense_2
    • Input: (500)
    • Output: (500)
  • Dense_3
    • Input: (500)
    • Output: (15)
• Dropout Layer Types
  • Dropout_1
    • Input: (500)
    • Output: (500)
  • Dropout_2
    • Input: (500)
    • Output: (500)
• Flatten Layer Types
  • Flatten_1
    • Input: (7,7,64)
    • Output: (3136)

This is the structure of the different layer types with their input and output. We hope that you are getting the points as of now listed. In this regard, we want to list out the CNN methods for your better understanding. Are you ready to know about that? Come let’s have the quick insights. In fact, we are going to illustrate to you the 2 of them.

• CNN 
  • It recognizes the non-trained classes of the finger vein images
  • Subject to the complexities compared to other methods
• CNN Reduced Complexity Four Layer
  • It doesn’t require a specific process for reducing dimensions & feature extraction
  • It can’t recognize the non-trained classes of the finger vein images

The above listed are the 2 major methods used for finger vein recognition. Apart from this, there are numerous methods are practiced. If you do want more details in these areas you are always welcome to have our suggestions. It is also important to know how to modify the structure of CNN for better performance. Yes, you people guessed right! We are going to show you the same for the ease of your understanding. 

How to Modify the Structure of CNN for Better Performance? 

• Convolutional Layer
  • Inception Module
  • Network in Network
  • Transposed Convolution
  • Tiled Convolution
  • Dilated Convolution
• Pooling Layer 
  • Multi-Scale Order less Pooling
  • Spatial Pyramid Pooling
  • Spectral Pooling
  • Stochastic Pooling
  • Mixed Pooling
  • LP Pooling
• Activation Functions
  • Probout
  • Maxout
  • ELU
  • RReLU & PReLU
  • LReLU & ReLU
• Loss Functions
  • KL Divergence
  • Triplet Loss
  • Softmax Loss
  • Hinge Loss
  • Contrastive Loss
• Normalization
  • Drop Connect
  • LP Norms
  • Dropout
• Optimization
  • Shortcut Connections
  • Batch Regularization
  • Weight Initialization
  • Data Augmentation
  • SGD
• Speed in Processing
  • Sparse Convolution
  • Weight Compression
  • Structured Transforms
  • Low Precision
  • FFT

The above listed are the layers in which we can modify the structure of CNN. We hope that you understand the concepts as of now listed. In addition to that, we also wanted to list out the latest CNN architecture for finger vein biometric identification. Next, we can have that section!!!

Latest Architecture for Finger Vein Biometric Identification using Convolutional Neural Network

• “Width” based Multi Connection CNN
  • Xception
  • ResNet
  • Inception Family
  • Pyramidal Net
  • Wide ResNet
• “Attention” based CNN
  • Concurrent Excitation & Squeeze
  • Residual Attention Neural Network
  • Convolutional Block Attention
• “Channel Exploitation” based CNN
  • TL Channel Boosted CNN
• “Feature Map Exploitation” based CNN
  • CompetitiveExcitation & Squeeze
  • Excitation & Squeeze
• “Multi-Path” based CNN
  • DenseNet
  • Highway Nets
• “Depth” based CNN
  • Inception ResNet
  • V3 & V4 Inception
•  “Spatial Exploitation” based CNN
  • GoogleNet
  • VGG & ZfNet
  • AlexNet

The foregoing passage has revealed to you the latest architecture components which are predominant to the CNN. Dataset is paying the dominant role in identifying the finger vein biometrically. Yes, the next section is all about the datasets used so far. 

Datasets for Finger Vein Biometric Identification

• IDIAP Research Institute VERA Finger Vein Database
  • VERA is a freely available open-source dataset
  • It contains the images of 440 from 110 clients
  • It also consisted of the spoofing cyber-attack info
  • This dataset helps to investigate the loopholes of the biometric devices
• University of Twenty Finger Vascular Pattern Database
  • UTFVP is also an open-source dataset of finger veins
  • It consists of 6 fingers from 60 volunteers
  • 6 fingers are the ring, middle & index from both hands
  • It is obtained by the university students via two sessions
• PLUSVein-FV3 Finger Vein Dataset
  • This is also a freely available dataset of human finger veins
  • It consists of both dorsal & palmar images from 60 subjects & 360 fingers
  • It considers the finger veins from both hands ring, middle & index fingers
  • These were obtained through a single session from 2 sensor variants
  • For illumination NIR lasers & for samples NIR LED
• MMCBNU_6000
  • It is also an open-source database of finger veins
  • It acquires the fingers veins from the 100 volunteers 6 fingers
  • The 6 fingers include index, ring, and middle of the 2 hands

The aforementioned are the major datasets used for finger vein recognition using Convolutional Neural Network. In fact, we are proficient with these datasets. By offering the various projects based on this we know the application areas exactly. We hope that you are enjoying this article to the core. It is always necessary to evaluate the performance of the finger vein biometric identification by several metrics. Come let’s also cover that section for your better understanding.

Performance Evaluation Metrics for Finger Vein Biometric Identification

Performance evaluation is comprised of several terms. We thought that it would be better to highlight them first.

• EER- Equal Error Rate
• FAR– False Accept Rate
• FRR– False Reject Rate

These are the terms indicated as the performance metrics of the finger vein biometric identification. Let’s have the importance of these metrics in the immediate section with clear points.

• EER handles the imposter & fine matching cases acquired from the vein recognition
• Matching cases always depends upon the matching scores
• Genuine matching cases are indicated as FAR
• Imposter matching cases are indicated as FRR
• Threshold point is acquired when both the FRR & FAR same

So far we have discussed and brainstormed in the areas of finger vein biometric identification using convolutional neural network. This article is represented the essential concepts comprised in the finger vein biometric identification. This technology has wide areas to perform. We hope that you would have benefited from this article and get interested to explore it further.

“Stay updated and keeps on exploring the technology to make you wise”