Alex Feldman

Alex Feldman

Project title: Improving fingerprint interoperability with Fourier domain texture descriptors

Project description: Optical fingerprint sensors are commonly used for identity authentication, but current systems perform poorly when many different sensors are used. By considering new fingerprint features which are robust to inter-device differences, we hope to improve accuracy when traditional matching techniques alone are not sufficient. Local texture descriptors may be the right tools for authentication even in cases with high device diversity.

Blog:

July 21

With one week left in the REU it is undoubtably time to reflect, and there is plenty to talk about. I presented much more over the summer than I imagined I would and rather unexpectedly, that was what I liked to do best. Learning new techniques is interesting and applying them to a novel problem is great, but working out how I can tell a unified story about the process is just more fun. When presenting my research I need to tie each element together and tell a compelling and linear account that is the fairytale version of what I actually did. I think about this when I pick a buzzword which hides my incomplete understanding of an equation or when I choose 3 rows of results to report and pretend the 200 “failures” never happened. Presentations need to relay interesting information, not represent the daily grind of research. Learning that out has been a lot of fun. I’m grateful to my advisor Ema Marasco for her guidance.

On the flip side, I do struggle to understand concepts and the overwhelming majority of my trials do fall short, which isn’t always easy to deal with. It’s been challenging to get these incremental results and feel satisfied day after day. There is what to learn from our research and I do expect it to make a difference somewhere down the line, but it feels very small to me. Focusing on other aspects of the REU, like presentation, talking to students in other fields, and considering my options after graduation have been important ways for me to cope with the disappointment.

I am looking forward to the program conclusion and poster session next week!

July 14

I felt as though the REU was really wrapping up this week. I spent most of my time rerunning experiments and finalizing my results. It’s been rather disappointing to see that the approach I was trying is not panning out as well as all of us in the lab had hoped. The focus is now on our poster session and I have plenty to present, but knowing that my path to solving an important problem made a difference would have been great. We are still considering ways to present our findings and may get accepted to a conference or workshop. That would be an amazing way to finish up a summer’s work.

Looking forward to what is yet to come.

July 7

“So what are you working on?” Even before getting to Charlotte, I’ve needed an answer to this surprisingly tricky question. What I’ve learned this week is that there is not a single answer, but each response must be tailored to the situation.

Because of the holiday, I haven’t spent much time in the lab this week, but have been talking to many friends who are all curious about my work. On top of that, the whole CCI REU gave short research presentations to peers and faculty this week where we needed to touch more deeply on the specifics of our research challenges and solutions. Coming from a third angle, I’ve been reading new research on texture descriptors and have needed to “explain” the concepts (to myself) as deeply as possible for my own understanding.

Sharing is what sets academic research apart from industry. Without journals or conferences, all knowledge would be stuck in a single place. I need to take the ways I present my work very seriously and treat each inquiry with respect. No matter the level, when discussing research the focus must remain on teaching to the audience above all else. With more practice I hope to grow in my ability to present to all.

June 30

Over the past few weeks, I’ve made many small mistakes in my work and have needed to find workarounds. Above all, that process has taught me to stay organized.

When working with large datasets, every step needs to be formalized algorithmically. This can be a very slow process, but it ensures that results and correct. Being confident in my calculations is very important because tracing back where a number originated can be painstaking work. There have also been a number of larger mistakes with the data, but organization helps there too. This week, we realized that the very first step of our process should have been image cropping. Hopefully our results will improve following the cropping, but it also means that all of our previous tests are now incorrect and need to be rerun. Thankfully, because our process is extremely systematic, the biggest loss will in computation time. This reflects a lot of what I’ve learned through my research more generally. I wind up repeating work time and again. It takes many hypotheses and tests to make any forward progress. The next time through, though, goes much more smoothly. Organization is key.

June 23

In our fourth week we’re finally getting publishable results. There’s certainly more testing to be done, but everyone is sure that we’ll have some novel research. That part feels very good, especially because we were unsure about our techniques in the first place. For now, I’ll keep running my models.

We’ve also given more presentations and thought about how to present the material to audiences of many different levels. After first understanding the topics myself, I need to present it to my advisor. From there, I have been working to distill it for experts image recognition, machine learning, or biometrics. Finally, in our final posters and presentations we are considering a high school level audience. Thinking about the work from all of these perspectives is quite challenging. I keep wanting to demonstrate expertise and use precise language, but when that isn’t how the audience will understand, I know I need to change my methods.

June 16

My project has really started getting going this week. After learning more about Random Forests and Support Vector Machines, I’ve started running models and finally getting results. I still feel I’m missing a lot of the advanced tools which would really set this work apart, and that’s frustrating. I know that for all the work I do in feature extraction and model creation, there are newer techniques which would outperform the things I’m trying. I’m getting helpful guidance from others in the lab and learning so much, but in two months I don’t know how I could be familiar with the state of the art in all relevant fields. Skills gaps notwithstanding, this has been a very good week in the lab. Seeing results come in is great and the immediacy of feedback is encouraging. After changing a few parameters in a model, I get results in a few minutes on how they perform and can continue tuning from there. I really hope to find an interesting modeling method from this. Support from the lab has been great. Each week I give a presentation to my advisor and some grad students who critique the content and the presentation itself which is usually more valuable. Most of the time I’m focused on content, but in those presentations, I understand how important a good pitch is. After working for the full 9 weeks, I’m sure the final presentation will go very smoothly.

June 9

This week I looked more closely at local fingerprint patterns and begin building a model to predict the authenticity of fingerprints considering their textures. I gave a presentation on the topic to my mentor and some grad students and their feedback was very helpful. They gave guidance on the little things like including slide number to more substantive issues about explaining the problem clearly. Overall, I learned a lot from putting together the presentation and from their feedback. Next week I’ll move on with our model and optimize the features we use in making predictions. First, I’ll need to read a lot more documentation on MATLAB random forest models. We also took part in a few seminars and activities. I appreciated hearing from the Associate Dean of graduate studies who gave guidance on applying to grad school. I also liked the low ropes course that we took part in today with the chemistry REU students. Looking forward to more in the weeks ahead!

June 2

Two months is an interesting timeframe to work on a research project. While it’s hard to see a project from start to finish in that time, it is long enough, I hope, to be immersed in a project and experience a research environment. This week I started that process by getting up to speed on my project. To improve interoperability of fingerprint recognition systems – that is how well fingerprints acquired on one scanner can be used by a different scanner – we are considering a number of features not previously employed for fingerprint analysis including a local descriptor of texture in an image known as local phase quantization (LBQ). I’ve spent most of my working days so far reading articles on fingerprint authentication, Fourier transformations, and local image descriptors. Each new concept has sent me searching for a more complex understanding of the topic in hopes of understanding the mechanisms at work. Today (6/2), I spent the day summarizing a paper on LBQ which has been a very challenging way to ensure I truly understand the descriptor. Even in this short time, I have made measurable progress in my understanding of a complex concept and appreciate the exercise of summarizing a challenging paper. Next week we’ll be implementing code to extract this features and others and using them to train a discriminator. I’m looking forward to moving on with the project in that way.