Centre for Medical Image Computing (CMIC)


Loic Peter - Sepehr Jalali

Start: Oct 04, 2017 01:00 PM
End: Oct 04, 2017 02:00 PM

Location: UCL Bloomsbury - Roberts 508 - Roberts building

Sepehr Jalali

Title: Through the Eye, to the Eye


In this talk, I will give a brief review of my journey from simulating human visual cortex, to getting inspirations from Dolphins echolocation, to sports psychophysics, to image guided-robotic retinal surgery. I will briefly introduce human visual cortex, models inspired from it and my contributions. I will talk about the way computer vision can be used in psychophysics and sports, and will finish this presentation with introducing my current challenge in OCT and retinal image enhancement, classification, segmentation and approaches to address it.

About the speaker:  Dr Sepehr Jalali is a post-doctoral research associate with the UCL institute of Ophthalmology and the Centre for Medical Image Computing, and a visiting lecturer of human and computer vision with City, University of London. He pursued his PhD and first post-doc at National University of Singapore.

Loic Peter

Title: Assisting Interactive Image Segmentation with Decision Forests


I will first present some results from my PhD research at the Technical University of Munich about the use of decision forests for semantic image segmentation. More precisely, we will see how pre-trained forest models can assist the interactive delineation of images in two application cases where standard techniques (such as graph-cuts, random walks...) are not directly applicable:

(i) the examination of very large digital slides in histopathology, where finding the objects to annotate is often the practical bottleneck for a clinician, and

(ii) the case of hands-free interactions, where a user can only communicate with the computer via binary "Yes/No" answers.

In addition, I will present a quick overview of my current research at UCL about image mosaicking from fetoscopy sequences, aiming at creating a full map of the placenta from a series of images captured with a very limited field of view.