In this blog post, I sit down with Raul Garcia-Martin, a PhD candidate in Biometric Recognition at the University Carlos III of Madrid.
Raul’s work focuses on identifying individual people by their biometrics. You’re likely already familiar with the most popular methods for biometric recognition:
- Face recognition
- Fingerprint recognition
- Retinal scans
… but did you know that the veins in your body can also be used for human identification?
This type of biometric recognition is called Vein or Vascular Biometric Recognition (VBR).
It’s not as well researched as other biometric recognition systems, but research shows that it can be just as accurate, if not more accurate, than the other methods.
Raul has been studying VBR throughout his graduate school career. As his Google Scholar profile shows, he’s already had numerous publications in this sub-niche of computer vision.
I have to admit that Raul reminds me a lot of myself when I was in graduate school.
Not only is Raul performing research and finishing up his PhD, but he’s also an entrepreneur. His company develops specialized cameras (and associated software) that can be used for infrared and thermal imaging.
These cameras can be used for:
- Fire detection
- Industry inspections
- Security
- Military applications
- COVID-19 body temperature detection
- … and not to mention, vein recognition!
To learn more about Raul’s work in Vein/Vascular Biometric Recognition, including how his work has helped him build his company, be sure to give the full interview a read!
An interview with Raul Garcia-Martin, PhD candidate and computer vision entrepreneur
Adrian: Hi Raul! Thank you for taking the time to do this interview. It’s a pleasure to have you on the PyImageSearch blog.
Raul: Hi Adrian! Thank you very much for giving me this great opportunity: you can’t imagine what it means to me to be here, it is an honor and a pleasure to contribute to the PyImageSearch blog. I sincerely hope that the PyImageSearchers could find something interesting in this interview that motivates them, as you do with me, to follow their dreams and the path to become computer vision and deep learning experts.
Adrian: Before we get started, can you tell us a bit about yourself? You’re a PhD candidate at the University Carlos III of Madrid, in addition to carrying out other projects, correct?
Raul: I am in my third year, out of four, as a PhD candidate in Biometric Recognition at the University Carlos III of Madrid. As you mentioned, at the same time, I am trying to go ahead with an entrepreneurial computer vision project.
Adrian: What got you interested in studying computer vision?
Raul: Since I was a child, I remember that I wanted to be an inventor. Around 2002, computer vision and deep learning weren’t as advanced as nowadays. I didn’t know anything about them, but my dream was clear: to engineer and develop technological solutions to improve people’s lives.
Therefore, with this clear yet undefined goal, and thanks to the infinite fields of knowledge that technology offers us, I studied for a bachelor’s degree in Industrial Electronics and Automation Engineering. In this sense, I feel very fortunate because I have had the opportunity to go to university, and I have always had the support of my family.
Without having found my way yet, I started a master’s degree in Electronic Systems and Applications. I combined both university studies with my first jobs as an electronics hardware and firmware developer in a small company in the industry sector and as a software tester in a multinational company in the railway sector.
But it wasn’t until I started my MSc thesis that I fell in love with computer vision. This occurred when I managed to get, for the first time, a video stream in real-time from a webcam using Python and OpenCV.
Adrian: Based on your Google Scholar profile, most of your work involves vein recognition and vascular biometric analysis using computer vision. Can you tell us a bit more about this research?
Raul: My PhD mainly addresses Vein or Vascular Biometric Recognition (VBR). It is a not very well-known biometric modality that uses the extraction and classification of unique human patterns to authenticate or identify people, just like facial or fingerprint recognition do.
There are four main VBR variants: finger, palm, hand dorsal, and wrist. I am researching wrist VBR because there are already patents and some commercial systems for the finger and palm vein modalities. Furthermore, I think wrist vein patterns are easier to visualize and capture.
Adrian: You recently published a paper in IEEE Access, Deep Learning for Vein Biometric Recognition on a Smartphone (Figure 1). Can you tell us a bit more about this paper? And in a COVID/pandemic world, why would we want to recognize veins using smartphones?
Raul: First of all, I would like to mention that I am very grateful to you because most of the deep learning knowledge presented in this article is based on the teachings extracted from your excellent Deep Learning for Computer Vision with Python book. I had no previous idea about deep learning (Convolutional Neural Networks, CNN, in this case), and in record time, I acquired solid knowledge with well-organized and structured information.
The main goal of this work is to bring Vein Biometric Recognition closer to our daily life, embedding this biometric variant into the small but powerful computer that has become an extension of our bodies: the smartphone. For this purpose, a deep learning model has been integrated into a smartphone for real-time video stream authentication and identification.
PyImageSearch readers can find a good video summary and demonstration here:
This authentication variant on smartphones, I think, could be a really interesting online payment or bank transaction method, being a comfortable and more secure alternative to facial or fingerprint verification.
In an attempt to contribute to this COVID/pandemic world, the other goal of this study is to develop vascular contactless multi-user devices. It is a more challenging computer vision technique, but it is optimal to prevent physical contact between the user and the device, providing a hygienic method of massive access control (e.g., airport border control).
In this sense, smartphones are portable and more comfortable devices that could be used, for instance, by an entrance operator in a sports stadium. In addition, this biometric variant is a secure alternative and more respectful in terms of user privacy than facial recognition, which nowadays presents an added challenge with masks.
Adrian: What type of hardware is needed to perform vein recognition? Is a standard iPhone/Android camera sufficient? Or do you need something more? I think this question will allow us to learn more about your passion: your computer vision entrepreneurship path.
Raul: Excellent question! To perform vein recognition, we only need a near-infrared camera (known as an IR camera). So answering your second question, a standard iPhone/Android or webcam is not sufficient. The RGB (standard) and IR cameras use the same sensor sensitive to visible radiation. But they physically mount, respectively, an IR blocking filter to render what for our eyes is a real image and a visible blocking filter (using the suitable IR torch) to see in the dark, through some plastics or visualize vein patterns.
I am completely in love with IR cameras.
Unfortunately (or not), it is not easy for us to access this type of camera on a smartphone because it is frequently used for facial authentication. However, that is why I love trying to access them. It is one of the reasons behind my entrepreneurship project, RGM Vision, where I’m “bringing infrared cameras and computer vision to people’s daily lives.”
On my website, PyImageSearch readers can find more than 5 infrared camera apps (Figure 3) for 6 different Android devices.
Furthermore, I have launched a new thermal camera available for all Android devices in my latest app: RGMVision ThermalCAM 1 (Figure 4).
I know that PyImageSearchers are more than acquainted with this type of IR camera (based on middle-infrared and far-infrared light). Still, if someone is interested, I recommend the latest interview on the PyImageSearch blog with Askat Kuzdeuov.
Adrian: What types of image preprocessing steps are required to perform vein recognition? Can you take the images directly from the infrared camera and apply deep learning models to them? Or do these images require additional preprocessing?
Raul: To perform vein recognition, near-infrared vein images are processed as grayscale images.
Before the eruption of deep learning in recent years, the first step was to preprocess the vein images to increase the contrast between the vascular patterns and the surrounding living tissue. Then, following the traditional biometric recognition paradigm, unique features were extracted and classified.
As we know, deep learning has changed this research methodology.
In this work, I have obtained good results both for preprocessed images, increasing the vein visualization using Contrast Limited Adaptive Histogram Equalization (CLAHE, Figure 2), and for raw grayscale images. The biometric recognition performance, using deep learning models, has not been substantially influenced by the preprocessing step. This seems to indicate that CNNs suffice to extract all relevant features.
Adrian: How did you settle on a model architecture for this project? Did you hand-design the model, or did you apply fine-tuning/transfer learning using existing architectures?
Raul: My first idea was to settle a state-of-the-art CNN architecture and train it from scratch. But in Deep Learning for Computer Vision with Python, I learned an even easier way: transfer learning. So, I implemented and tested both transfer learning variants, i.e., CNN as feature extractors and fine-tuning, over state-of-the-art CNN pre-trained architectures.
For the UC3M-CV2 dataset, only the CNN as a feature extractor technique obtained a high accuracy, as it could be expected, according to your advice, when we work with a small dataset.
Adrian: What computer vision and deep learning tools, libraries, and packages did you utilize in your research?
Raul: Since I started my journey in computer vision, I have always used OpenCV due to its ease of use and simplicity. I love this library! I program using Python for the same reason, even though I hadn’t tried it before starting with computer vision.
Since I began to apply deep learning, I have been using Keras and TensorFlow, but maybe I should also catch up with your latest PyTorch tutorials.
Other than that, as PyImageSearch readers will have already discovered, I love programming with Android: for me, it is the most ready-to-produce programming language.
Adrian: What are your next steps as a researcher? Are you going to continue working on automatic vein recognition, or are you moving on to other projects?
Raul: My idea is to finish my PhD in vein recognition and then try to make a living from computer vision and camera development: I’ll put all my efforts into building a thriving business.
At this point, I would like to change the roles of this interview and ask you for advice on how to become a successful entrepreneur and launch a startup/business (I know that it is related to one of the possible points of your brand new set of courses, “A technical education is not enough to succeed and hit your goals”).
I know this path is hard, but I believe I must try it. In addition, I keep the doors open to any other possibility.
Adrian: Congrats on being one step closer to completing your PhD! What advice would you give to someone who wants to follow in your footsteps, complete their PhD, and become a successfully published researcher?
Raul: Thank you very much, Adrian!
I recommend that they be persistent (in all aspects of their lives, anything worthwhile in our existence needs it) and enjoy the journey putting all their passion into their research. Results will come along.
As practical advice, I have to mention that I have always been a great fan of self-taught learning. In my case, I think that the university has prepared me for it. I also believe in another powerful concept: always remain curious and learn as much as you can from everybody. Especially from the main references in every field without fear of investing in yourself and your education.
Adrian: You’ve been a long-time reader and customer of PyImageSearch. Thank you for supporting us! How has PyImageSearch helped you with your research and journey to completing your PhD?
Raul: I am indebted to you for your teachings and inspiration!
Through this interview, I’ve already mentioned that most of my deep learning knowledge comes from your Deep Learning for Computer Vision with Python.
And like everybody else in the computer vision world with Python, I have been learning throughout these years (and currently keep doing so) from your over 350 free tutorials. Thank you again!
If I had to start over again from scratch in computer vision, I would begin without hesitation by checking out your books. At this point, I think that PyImageSearch University could be my next step.
Adrian: If a PyImageSearch reader wants to connect with you, how can they do so?
Raul: It would be a pleasure if any PyImageSearch reader wants to connect with me!
They can contact me on LinkedIn or through the RGMVision website via WhatsApp or Telegram.
Summary
Today we interviewed Raul Garcia-Martin, PhD candidate at University Carlos III of Madrid specializing in computer vision and Vein/Vascular Biometric Recognition.
Raul’s work shows that it’s possible to identify a person using the veins in their body, specifically the hands, wrist, and forearms. Furthermore, this method is just as accurate, if not more accurate, than other biometric recognition methods (i.e., face recognition, fingerprint recognition, etc.).
Additionally, Raul is developing his own set of cameras and software to facilitate further research in this area. Be sure to check out his company, RGM Vision, for more information.
To be notified when future tutorials and interviews are published here on PyImageSearch, simply enter your email address in the form below!
Join the PyImageSearch Newsletter and Grab My FREE 17-page Resource Guide PDF
Enter your email address below to join the PyImageSearch Newsletter and download my FREE 17-page Resource Guide PDF on Computer Vision, OpenCV, and Deep Learning.
Comment section
Hey, Adrian Rosebrock here, author and creator of PyImageSearch. While I love hearing from readers, a couple years ago I made the tough decision to no longer offer 1:1 help over blog post comments.
At the time I was receiving 200+ emails per day and another 100+ blog post comments. I simply did not have the time to moderate and respond to them all, and the sheer volume of requests was taking a toll on me.
Instead, my goal is to do the most good for the computer vision, deep learning, and OpenCV community at large by focusing my time on authoring high-quality blog posts, tutorials, and books/courses.
If you need help learning computer vision and deep learning, I suggest you refer to my full catalog of books and courses — they have helped tens of thousands of developers, students, and researchers just like yourself learn Computer Vision, Deep Learning, and OpenCV.
Click here to browse my full catalog.