The development of efficient algorithms—whether in terms of execution speed, memory footprint, or ease of use—is at the core of the team’s work. This focus is reflected in several recent projects, particularly in the field of image processing.

Our contributions in high-performance image processing

The team actively contributes to the development of open-source libraries that provide reusable and efficient image processing tools. We recently addressed the following challenges:

• Developing high-performance mathematical morphology algorithms that exploit massively parallel architectures (e.g. GPUs). We focus on the porting of some tree-based image representations widely used in mathematical morphology (the max-tree, the tree of shapes and the α-tree). Our algorithms achieve the state-of-the-art performance in this field [2].

• Providing a mathematical morphology-related Modern C++ library that conciliates ease-of-use, performance and genericity. Pylene aims at providing mostly Mathematical Morphology building blocs for image processing pipelines [1][3][4]. It features the following characteristics:
  • Simplicity it has both python bindings and simple C++ syntax
  • Efficiency with algorithms written in a simpler language that run as if they were written in C. We follow one guideline: zero-cost abstraction.
  • Genericity with the abaility to write algorithms once that run on many kind of images with zero-cost abstraction.
  • Interopability with algorithms that run on image coming from external libraries (even on your own image type).

A single code (algorithm) for many types. A watershed on 2D images, meshes and 3D volumes.

Related Publications