Talk by Raffi Enficiaud: Generic approaches for algorithms in mathematical morphology
Abstract
A wide class of problems and multiple type of sensors (2D, 3D, 3D+t, color, tensorial… ) exist in computer vision. Despite this diversity, the involved algorithms often use the same elementary tools with slight variations that are dependent on the application. Even small, these variations make the algorithms strongly coupled with specific needs. This led to hardly reusable software and developments that can be in most cases avoided, which indirectly impacts the time spent on research.
In this talk, we present the paradigms that allow genericity and reusability of computer vision algorithms with an emphasis on mathematical morphology. The genericty is achieved by abstracting the main mathematical notions, and by splitting complex algorithms into reusable components. As a principal result of this approach, the prototyping of new image analysis algorithms at reduced development costs is made possible. As a secondary effect, classical morphological algorithms become easily multidimensional and multispectral. We illustrate our presentation in several concrete applications and focus on morphological segmentation and color processing.
Short Bio
Raffi Enficiaud is a research engineer at INRIA Paris-Rocquencourt. He works in the IMEDIA team on Content Based Image Retrieval. He received the engineering degree from the National Institute of Telecommunications, France, in 2000. After working as a consultant, he became a graduate of the Mathematical Morphology Center of the Paris school of Mines, where he obtained his Ph.D. in computer science in 2007. He then worked as a research engineer on image quality evaluation at
DxO? Labs, Paris. He joined INRIA during 2008. His fields of interest include image analysis (mathematical morphology, segmentation, color and feature extraction), machine learning and information visualization.
