Fachbereich Informatik - Aktuell


Einladung zum Vortrag von Prof. Dr. Michael Werman (Hebr. Universität Jerusalem)

am Dienstag, den 2. Oktober 2018, um 14:15 Uhr in Raum C424, Sand 14

Image declipping with deep networks

We present a deep network to recover pixel values lost to clipping. The clipped area of the image is typically a uniform area of minimum or maximum brightness, losing image detail and color fidelity. Clipping may occur in any (or all) of the pixel's color channels and occur to some degree in almost every image we tested.,  Using neural networks and their ability to model natural images allows our neural network, DeclipNet, to reconstruct data in clipped regions producing state of the art results.

Sketch-based reduced memory hough transform 

This paper proposes using sketch algorithms to represent the votes in Hough transforms.Replacing the accumulator array with a sketch (Sketch Hough Transform - SHT) significantly reduces the memory needed to compute a Hough transform.

Two view constraints on the epipoles from few correspondences

In general, it requires at least 7 point correspondences to compute the fundamental matrix between views. We use the cross ratio invariance between corresponding epipolar lines, stemming from epipolar line homography, to derive a simple formulationfor the relationship between epipoles and corresponding points. We show how it can be used to reduce the number of required points for the epipolar geometry when some information about the epipoles is available.