It's really interesting that it almost feels like you can "recover" information from the image just by applying this convolution. I guess in this case, though, blur isn't necessarily a lack of information, which in and of itself is weird to think about. Is this how photo-editing software like photoshop applies sharpening, or is something more robust used for that?
GuardHei
@showyouramen I feel the same. Blur doesn't mean a lack of information. A good example is anti-aliasing, where you "blur" at the geometry edge.
In terms sharpening, I think this is the easiest way of achieving this. However, having a negative lobe could also easily result in overshooting, which will cause ringing artifacts. AMD has developed an adaptive sharpening tech (F-CAS), which tries to only sharpen areas needed. There are other kernels to "sharpen" the images, including bicubic sampling, Catmull sampling, lanscoz sampling (used in AMD's FSR upscaling tech) and etc. More advanced methods will sharpen the pixel anisotropically (like FSR or Nvidia's NIS).
It's really interesting that it almost feels like you can "recover" information from the image just by applying this convolution. I guess in this case, though, blur isn't necessarily a lack of information, which in and of itself is weird to think about. Is this how photo-editing software like photoshop applies sharpening, or is something more robust used for that?
@showyouramen I feel the same. Blur doesn't mean a lack of information. A good example is anti-aliasing, where you "blur" at the geometry edge.
In terms sharpening, I think this is the easiest way of achieving this. However, having a negative lobe could also easily result in overshooting, which will cause ringing artifacts. AMD has developed an adaptive sharpening tech (F-CAS), which tries to only sharpen areas needed. There are other kernels to "sharpen" the images, including bicubic sampling, Catmull sampling, lanscoz sampling (used in AMD's FSR upscaling tech) and etc. More advanced methods will sharpen the pixel anisotropically (like FSR or Nvidia's NIS).