70 likes | 89 Views
Corrected Connected Component. Connected Component Algorithm. Two pass algorithm: assign labels and track equivalences reassign labels based on equivalences. When traversing the image, only the neighborhood pixels that have already been encountered(shown inyellow) are observed.
E N D
Connected Component Algorithm Two pass algorithm: • assign labels and track equivalences • reassign labels based on equivalences. When traversing the image, only the neighborhood pixels that have already been encountered(shown inyellow) are observed. An equivalence is made if any two yellow pixels have different labels.
Connected Component Algorithm What I did wrong: I had a problem with my equivalences. I was not updating them correctly. For example: The red component will be created when the 2nd row is reached. On the third row an equivalence will be made between green and red. green=red On the third row an equivalence will be made between blue and green blue=green When I relabeled the image I was not properly travelling through the equivalences. So blue would not always be made equivalent to red.
Corrected Results Incorrect Corrected Component Analysis Algorithm
Corrected Results Incorrect Corrected Algorithm The corrected algorithm presents a new problem. Because there are gaps in the contour of the face and hair the whole region will be connected. This prevents the dams around the face from being built properly because both sides of a ridge belong to the same basin.
Corrected Results Gaps in contour
Corrected Results • The next step is to fix the contour gaps. 2. Possible solutions: Use a different method to obtain the gradient that will keep edges more continuous Perform some operation to fill in the gaps