[WEEK 3] Image Colorization Results With Filters
Hello to everyone again. This week we will examine the importance of edge information for the colorization process with gaus filter effect. Then we’ll talk about how we plan to follow in the following weeks.
This week we tried to make colorization by reducing the edge information in the images . Our goal in doing this is to observe the reflexes that our trained network has given to the objects and to identify which objects are better recognized and more beautiful colorized. While doing this, we will apply the gaus filter we all know to our input images. Because we know that the gaus filter is a high pass filter that reduces the edge information in the image.
The results are pretty bad when the guaus filter is applied, as is evident in the inputs and outputs. Instead of recognizing the trees separately, they have known as big trees and missed the detail changes.
here, the bird’s wing in green color, the network thinking that the tree. Can not distinguish objects because edge information is lost. Therefore it will not be a good method to use the gaus filter. But as we mentioned in the previous article, when we used sharpen filter, we knew objects better and success was increasing. But there were different problems that caused the objects to know better.
In line with the results we obtained, we thought that what we actually needed could be a bilateral filter. Because it prevents the objects from detecting detail objects while preserving the edge information. We aim to increase the color differences between objects by providing color integrity within the object.
See you next week..
