Seite - 106 - in Proceedings - OAGM & ARW Joint Workshop 2016 on "Computer Vision and Robotics“

(a) test image (b) test image with Gaussian noise (SNR=3) (c) test image with Gaussian noise (SNR= 1.5) (d) reference (e)hitmap (SNR=3) (f)hitmap (SNR=1.5) Figure 1. Illustration of a pattern matching problem: find reference image in the test image. The images are taken from frame 697 and frame 705 of the EC Funded CAVIAR project/IST 2001 37540 (”Shopping Center in Portugal”,”OneLeaveShop2cor”). The hitmap iscomputed using the measure (1) ‖f‖D := max { max 0≤k≤N,0≤l≤M { k∑ i=0 l∑ j=0 I(i,j)}− min 0≤k≤N,0≤l≤M { k∑ i=0 l∑ j=0 I(i,j)} , max 0≤k≤N,0≤l≤M { k∑ i=0 l∑ j=0 I(N−i,j)}− min 0≤k≤N,0≤l≤M { k∑ i=0 l∑ j=0 I(N−i,j)}, max 0≤k≤N,0≤l≤M { k∑ i=0 l∑ j=0 I(i,M−j)}− min 0≤k≤N,0≤l≤M { k∑ i=0 l∑ j=0 I(i,M−j)}, max 0≤k≤N,0≤l≤M { k∑ i=0 l∑ j=0 I(N−i,M−j)}− min 0≤k≤N,0≤l≤M { k∑ i=0 l∑ j=0 I(N−i,M−j)} } (1) For registrationand templatematchingpurposes thediscrepancymeasure isappliedon thedifference of the corresponding images. To be more precise, the images are considered as two-dimensional functionson the latticeof integers withwithdefault values0 outside theproper frameof the images. Thismeasure satisfies the followingdesirable registration properties (seealso [8,9]): [R1] avanishingdistanceentails avanishingextentofmisalignmentand viceversa, 106