Subject

A lensless camera works by capturing light that passes through a diffusing material rather than a traditional lens, resulting in raw sensor data that appears unintelligible. Instead of relying on optics to focus light and form an image, a lensless camera uses computational imaging algorithms to reconstruct the image from the diffuse light patterns, addressing an inverse problem akin to those found in compressive sensing. This innovative approach allows for highly compact and lightweight camera designs, opening up various applications. In biomedical imaging, lensless cameras can be integrated into tiny endoscopic devices for minimally invasive procedures. Their small size and low profile in surveillance and security make them ideal for discreet monitoring.
Additionally, they hold promise in scientific research, such as microscopy, where traditional lens systems are bulky and limiting. Their potential for wide-field imaging and flexibility in design further extend their applications to areas like augmented reality (AR) and wearable technology, where integrating conventional lenses would be impractical.

Kind of work

For this thesis, the student will conduct a practical study of various reconstruction algorithms (task-agnostic) and detection algorithms (task-oriented), based on a real lensless camera prototype built at ETRO. Such algorithms will be designed using a machine learning approach and trained on a real dataset. Specifically, the emphasis will be put on low-complexity and interpretable deep learning methods, which have the advantage of requiring small training sets and being inexpensive during inference. The thesis will also focus on the real-time implementation of the chosen method on hardware (FPGA, Nvidia Jetson, …).

Framework of the Thesis

A lensless camera prototype is already developed at ETRO and a dataset will be already given to the student(s). The student will have the opportunity to investigate AI algorithms for camera data reconstruction and analysis and to develop them in real-life.

Sore useful references are given below:
https://opg.optica.org/optica/fulltext.cfm?uri=optica-5-1-1&id=380297
https://opg.optica.org/oe/fulltext.cfm?uri=oe-30-22-39520&id=509832
https://ieeexplore.ieee.org/stamp/stamp.jsp?tp=&arnumber=9239993

Number of Students

1 or 2

Expected Student Profile

Must-have skills: python programming, machine learning
Other useful skills: C programming, HDL and FPGA programming, image processing and computational imaging, electronics prototyping