Asgn 2, Problem 1

Moderator: Computer Vision

redDragonfly
Windoof-User
Windoof-User
Beiträge: 25
Registriert: 14. Apr 2010 22:32

Asgn 2, Problem 1

Beitrag von redDragonfly » 23. Mai 2012 20:06

Image Sharpening
We have difficulties understanding the final task of the first problem. The way it is written is very confusing concerning the order of action, so we'd like some confirmation if we understood the task correctly.
The task is:
(...) amplify the high-frequency components contained in the finest 2 levels of the Laplacian pyramid by scaling them up by a factor. Afterwards, reassemble the pyramid back to obtain a sharpened full-resolution image and display it.
First, "scale up" means "take the Laplacian images and multiply with some number", right? Otherwise, which purpose would scaling the image up (with our function upsample2) have?

Second, "reassemble the pyramid" does not refer to the construction of the Laplace pyramids, but rather to the reconstruction of the original image, is that correct? Only here we have to manually manipulate the last two steps of the reconstruction and multiply the Laplace images with some factor before adding the Gaussian image?

This leads to the final question: why the final two levels? Manipulating the finest level (L_0) already results in a discernible sharpening effects.

Toobee
Sonntagsinformatiker
Sonntagsinformatiker
Beiträge: 225
Registriert: 7. Apr 2011 12:58

Re: Asgn 2, Problem 1

Beitrag von Toobee » 24. Mai 2012 21:45

I am confused too regarding 'sharping with the finest laplace layers'.

Regarding the rebuilding: You shall rebuild the image from the laplace pyramide. It is expected that you already have the laplace for this task.
Accordingly to the slides you start with the lowest layer (8x8) scale it up add and it with next higher layer. You do that until you have added up all laplace layers with each other.
If you did it correctly, your resulting image will be the input image.

qgao
Moderator
Moderator
Beiträge: 33
Registriert: 7. Apr 2009 11:38

Re: Asgn 2, Problem 1

Beitrag von qgao » 29. Mai 2012 17:59

redDragonfly hat geschrieben: First, "scale up" means "take the Laplacian images and multiply with some number", right? Otherwise, which purpose would scaling the image up (with our function upsample2) have?
yes, you are right.

Second, "reassemble the pyramid" does not refer to the construction of the Laplace pyramids, but rather to the reconstruction of the original image, is that correct? Only here we have to manually manipulate the last two steps of the reconstruction and multiply the Laplace images with some factor before adding the Gaussian image?
After the manipulation of two finest scales, you will get a new Laplacian pyramid, with which you should then construct the corresponding image.

qgao
Moderator
Moderator
Beiträge: 33
Registriert: 7. Apr 2009 11:38

Re: Asgn 2, Problem 1

Beitrag von qgao » 29. Mai 2012 18:14

Toobee hat geschrieben:I am confused too regarding 'sharping with the finest laplace layers'.

Regarding the rebuilding: You shall rebuild the image from the laplace pyramide. It is expected that you already have the laplace for this task.
Accordingly to the slides you start with the lowest layer (8x8) scale it up add and it with next higher layer. You do that until you have added up all laplace layers with each other.
If you did it correctly, your resulting image will be the input image.
I think you misunderstood the term "scaling". In the problem it does not mean resizing of a matrix but the multiplication of the coefficients with some value.

Antworten

Zurück zu „Computer Vision“