Want to write your Master’s Thesis with us?

Project 1

Title: Outpainting for Synthetic Fingerprint Image Generation

Description: When training fingerprint recognition neural networks and developing algorithms for fingerprint applications, it would often be useful to work on larger images than one has access to. This could be solved by outpainting methods, that can extrapolate and create entirely new content outside the existing image. Models like GANs or DMs can learn to understand and replicate the patterns, textures, and structures present in the training data, enabling them to generate novel and contextually relevant content that seamlessly extends the original image. The thesis’s aim could be to generate a complete live fingerprint from a patch, or to enlarge an existing patch.

Project 2

Title: Improve Classification Robustness through Trained Augmentation

Description: CNNs are powerful tools for fingerprint matching and liveness classification. The key to success is training them on the right data. The aim of this thesis is to set up a network to augment images to make them harder for liveness or matcher classifiers. The augmentation can be used either to improve the classifier during training or to offline generate images for future training. In both cases, the ultimate objective is to make the classifier more resilient to variations in input data and to expose weaknesses in the input data.

Project 3

Title: Orientation Field Estimation in Fingerprint Images

Description: For optimal matching performance of fingerprint images, a robust orientation field estimation is crucial. The estimated orientation field should be robust to large amounts of noise, capturing artifacts, different image cropping, and so on. The master’s thesis could approach this with classic approaches, CNNs or a combination of methods.

Project 4

Title: Removal of Moiré Patterns in Fingerprint Images

Description: For under-display sensors a moiré pattern is overlayed on to the fingerprint image. The pattern is created by an alias between the display and the sensor pixels and is very characteristic. It changes with pressure and temperature, and it therefore becomes difficult to filter out, however the pattern has specific characteristics. The goal of this thesis is to extract and remove the Moiré pattern from the fingerprint using CNN.

Project 5

Title: Vision Transformers

Description: Vision Transformers (ViTs) are a novel architecture for computer vision tasks, inspired by the success of Transformer models in natural language processing. Unlike traditional convolutional neural networks (CNNs), which rely on convolutional layers to extract local patterns (e.g., edges, textures), ViTs apply self-attention mechanisms to process images globally.

In the master thesis we want to examine how vision transformers can be used for image matching. We have two different questions that can be examined:

• One is the use ViTs to generate a best possible alignment between two images. For example using ViTs as local descriptors
• The other is to use ViT to tell, given a best possible alignment whether or not two images match

Project 6 

Title: CNN based alignment methods

Description: A common computer vision task is to align (register) images for further processing. With image perturbations and low SNR conditions, this can become a tricky problem for traditional methods. In this thesis we want to investigate performance of CNN based methods (e.g. LIFT) for registering images.

Project 7 

Title: CNN quick match

Description: In a fingerprint matching scenario, one key task is to quickly find candidate matches in a large set of potential candidates. We want to explore CNN based methods, for instance, Visual Transformers, learned bag-of-features, or other similar representations, for quickly finding candidate matches for more thorough analsyis.

Project 8 

Title: Swipe Enroll

Description: When a user gets a new phone, they first need to enroll their fingerprints. This can be done in multiple ways, and an emerging approach is to let the user swipe the finger over the sensor. This creates multiple challenges for a fingerprint matcher, for example in handling distortion of the enrolled images. The aim of this thesis is to set up a system for getting high quality enrolled images out of a swipe. It can be approached with CNNs as well as classical approaches.

Project 9

Title: Generate large amounts of synthetic training data

Description: When training palmprint recognition neural networks and developing algorithms for palmprint applications, it is useful to be able to generate large amounts of synthetic training data. Precise Biometrics already has the capability of generating this kind of training data with certain limitations. This thesis’ aim is to generate groups of images as “genuines”, images that should match within the group, and have the similar characteristics, similarities and differences as images from the same palmprint in the real-world data.

Project 10

Title: Palm UI/UX

Description: Precise Biometrics, currently specializing in fingerprint solutions, is expanding into hand recognition technology. We are seeking to develop a comprehensive hand recognition solution integrated with our YOUNiQ Access system. A crucial aspect of this project is designing a user-friendly interface to ensure correct hand positioning.

Project Focus:
The contactless nature of hand recognition poses challenges in obtaining high-resolution, quality images without physical contact. This thesis will investigate optimal methods to guide users in using such a system.

Potential Research Areas:

• UI design for instructing proper hand positioning.
• Technical aspects of contactless hand recognition and image processing.
• User experience (UX) in relation to hand recognition.
• Evaluation of methods to teach users how to use the system.

Knowledge in UI/UX design and/or image processing is expected.

We look forward to working with a dedicated student to advance the future of technology in hand recognition and user-friendly interface design.

Take a look at previous thesis projects at Precise:

• Fingerprint Synthesis Using Deep Generative Models: https://precisebiometrics.com/blog/meet-our-masters-thesis-students-synthesizing-fingerprint-images/

• Pattern-based palm recognition for identification system using deep learning: https://precisebiometrics.com/blog/meet-our-masters-thesis-students-pattern-based-palm-recognition/

• Detecting Fingerprint Images Outside Training Distribution for Spoof Detection: https://precisebiometrics.com/blog/meet-our-masters-thesis-students-out-of-distribution-detection/