len0x
3rd October 2004, 22:26
Usually it's a known fact that compressibility of resizers goes like this:
bilinear > soft bicubic > neutral bicubic > sharp bicubic > lancsoz > lanczos4
However I came across one sample that produces this (in brackets there are relative sizes):
soft bicubic(100%) > bilinear (100.5%) > lanczos4 (103.6%) > lanczos (106.3) > neutral bicubic (107.2) > Sharp Bicubic (108.1%)
How come? If you don't take into account small difference on a given source between soft bicubic and bilinear there is one big puzzle for me: why lanczos produces much softer image then Bicubic ???
P.S. The only unusual thing about the source that its interlaced and in the script KernelDeInt 1.5.1 was used. If I use any other deinterlacers then first formula always apply. So what so special about lanczos - is it not guaranteed to produce sharper image ?
bilinear > soft bicubic > neutral bicubic > sharp bicubic > lancsoz > lanczos4
However I came across one sample that produces this (in brackets there are relative sizes):
soft bicubic(100%) > bilinear (100.5%) > lanczos4 (103.6%) > lanczos (106.3) > neutral bicubic (107.2) > Sharp Bicubic (108.1%)
How come? If you don't take into account small difference on a given source between soft bicubic and bilinear there is one big puzzle for me: why lanczos produces much softer image then Bicubic ???
P.S. The only unusual thing about the source that its interlaced and in the script KernelDeInt 1.5.1 was used. If I use any other deinterlacers then first formula always apply. So what so special about lanczos - is it not guaranteed to produce sharper image ?