It's interesting how one is able to increase the detail of an image without any additional information. Does this require some edge detection under the hood?
aramk-hub
I think this is a good point to the above comment. How exactly without taking another picture can a phone or camera adjust the sharpness? It can't see the edges. Does the camera just create darker shadows to outline the bars on the bridge for example? What is the operation that is done?
Dezhang1999
@smsunarto I was wondering the same, I think it just uses what it already had, and extend it or uses some algorithm to produce more information for this to happen.
somaniarushi
In addition to the great questions above, I also noticed that the second picture was just the slightest bit more noisy than the first one. What is the reason behind this?
stephen422
I'm thinking similar to what we do in JPEG, we can do DCT, but when quantizing, bump up the high frequency values rather than killing them to zero. That way we can amplify high frequency signal to the image and re-introduce sharp edges, albeit at the same time amplifying high freq errors as well.
seenumadhavan
One approach shown later in the slides is to use the sharpening filter.
shreyaskompalli
Throughout this class, I've found it really intriguing that so many of these complex techniques boil down to just matrix multiplications at the end of the day. Filters, transforms, and color computations all consist of just matrix products in the end, and its interesting to finally make the connection between the linear algebra that I've learned for years and some practical, visual applications.
It's interesting how one is able to increase the detail of an image without any additional information. Does this require some edge detection under the hood?
I think this is a good point to the above comment. How exactly without taking another picture can a phone or camera adjust the sharpness? It can't see the edges. Does the camera just create darker shadows to outline the bars on the bridge for example? What is the operation that is done?
@smsunarto I was wondering the same, I think it just uses what it already had, and extend it or uses some algorithm to produce more information for this to happen.
In addition to the great questions above, I also noticed that the second picture was just the slightest bit more noisy than the first one. What is the reason behind this?
I'm thinking similar to what we do in JPEG, we can do DCT, but when quantizing, bump up the high frequency values rather than killing them to zero. That way we can amplify high frequency signal to the image and re-introduce sharp edges, albeit at the same time amplifying high freq errors as well.
One approach shown later in the slides is to use the sharpening filter.
Throughout this class, I've found it really intriguing that so many of these complex techniques boil down to just matrix multiplications at the end of the day. Filters, transforms, and color computations all consist of just matrix products in the end, and its interesting to finally make the connection between the linear algebra that I've learned for years and some practical, visual applications.