Wednesday, 23 August 2017

PhD topics

From Image

Jump to: navigation, search

Remark: these subjects are foreseen research topics proposed by the lab staff (no funding is currently associated).


GPU-Computational 3D Light Field camera

Description :

Computational Imaging is an innovative technique where non-traditional image data sensing approaches (coded apertures, multi-aperture imaging, micro-lens arrays, sparse data mixing, …) provide multi-dimensional information (spatial, angular, colorimetric, …) that has to be computationally processed to fully comprehend the underlying 3D scene information. Such 3D Light Field cameras are the appropriate candidate for enabling 3D photo-realistic rendering (glasses-free 3D multiscopic displays, holography, 3D microscopy, …).

The PhD student will - in collaboration with external partners, who are experts in hardware imaging systems and mathematical signal conditioning - develop an integrated system with GPU-accelerated signal processing for 3D Light Field acquisition (with integrated depth estimation, light rays interpolation, …), covering specific 3D application demands. A strong background in image processing and C/C++ programming is a pre-requisite. Expertise in Light Fields is an added value.


Real-time MPEG Augmented Reality in Cultural Heritage

Description :

In the digital era, preservation of Cultural Heritage often involves intangible representations through digital archiving and augmented reality, allowing free viewpoint navigation and 3D visualizations through stereoscopic goggles (or immersive 3D displays).

The acquisition of the real scène - in which offline-prepared virtual objects have to be inserted - should be extended with a huge amount of real-time data processing (depth estimation for proper virtual object insertion in the real scène, environmental map recoloring, …) and coding steps (point cloud to 3D mesh coding, server-client progressive 2D texture coding, …) for fluent transmission and latency-free rendering in a server-client-based MPEG-ARAF framework (ARAF: Augmented Reality Application Framework within the MPEG standardization process).

The PhD student will develop and GPU-accelerate the core data processing required to capture, reformat and transmit data through an end-to-end MPEG-ARAF augmented reality application for architectural heritage. Proof-of-concepts using mostly existing tools in 3D acquisition and coding (in collaboration with internal/external experts in the field) will be built and refactored, eventually achieving the real-time processing constraints through a methodological approach in GPU and webGL programming.


Light Field Computational Imaging

Description :

Computational Imaging is an innovative imaging technique where computation plays an integral role in the image formation process, without which only illegible images would be acquired.

Though CMOS scaling has reached unpreceded levels, pixel scaling in CMOS sensors has limited future potential, especially in view of the ever-increasing demands in spatial resolution, high dynamic range and spectral/color separation (multi-spectral imaging).

New methods for image recovery from sparse and indirectly observed data (compressive sensing), as well as non-traditional image data sensing approaches (multi-aperture imaging, light field acquisition) have recently appeared as an effective way for overcoming the technology limitations with image processing techniques, though this often requires an inventive optical, CMOS sensor and algorithmic co-design.

The PhD candidate will closely co-operate both with imager design teams in IMEC and image processing labs in academia, following a multi-disciplinary approach for specific case studies in computational imaging applications (e.g. compressive sensing for multi-spectral light field acquisition in fluorescence bio-medical applications).


Multi-resolution biomedical image registration

Description :

The analysis of the tissue at the microscopic level is essential to validate the relevance of a radiotracer or the effect of a treatment. The major difficulty is to be able to match the in vivo anatomical (MRI,CT) or functional (fMRI, PET, SPECT) images with histological images obtained using whole slide imaging (WSI), which is becoming the standard in Digital Pathology.

The project aims to map the data from these two levels of analysis (macro/micro). These developments will help to validate the relevance of new PET tracers and new functional imaging methods (fMRI with or without contrast agent) by correlating the macroscopic imaging in vivo with detailed analysis at histological level, notably through the use immunohistochemistry (IHC). IHC markers can target proteins with functions involved in tumorigenesis, the effect of anti-tumor treatments and the response of the host tissue, including specific markers of cell proliferation, death (apoptosis), hypoxia, angiogenesis and inflammation.


Automatic extraction of knowledge from microscopy image database

Description :
This research is done in the context of the development of a the image analysis pole of the CMMI (Center for Microscopy and Molecular Imaging). Since the quantity of generated images of the center is huge, we want to develop a research topic on the automatic extraction of knowledge from large collections of annotated and non-annotated images.