Page - 188 - in Proceedings - OAGM & ARW Joint Workshop 2016 on "Computer Vision and Robotics“

(a) Original500 seg. (b)Uniform noise500seg. (c)Gaussiannoise 500seg. (d)Poissonnoise 500seg. (e)Original 2000 seg. (f)Uniform noise2000 seg. (g) Gaussian noise 2000 seg. (h)Poissonnoise 2000seg. Figure4: Reconstructedoriginal andnoisy images (a)Original images (b) Images with uniformnoise (c) Images withGaussian noise (d) ImageswithPoissonnoise Figure5: GCEsegmentation error of the test imagesandnoisy images well as theSNR:highamountofdetails (lowSNR)andaverageamountofdetails (higherSNR).The lattergroupof images corresponds with thepreviously expected behavior. For images with Gaussian distributed noise (see Figure 6b), we conclude that the GCE is almost the same as for the original images with a few outliers at maximal difference of 0.2. The Poisson distributednoise(seeFigure6c)hasasimilarbehavior touniformlydistributednoisebut lessoutliers. 188