vade

Ive implemented polar to cartesian and cartesian to polar shaders 

(these work like the photoshop filters, c74s is a little odd on my 

system) that allow doing some nice blurring radially. However, I am 

getting a vertical line in my output. I know this is introduced by 

the polar conversion, and so I ask, is there another mapping or polar 

type mode that keeps the edges of the cartesian image in line with 

the edges of the polar image, so it does not introduce any odd 

artifacts? It makes a bit more sense when you see an image.

Here are the shaders and a patch (these will be part of a release, so 

http://www.abstrakt.vade.info/ILOVETHEMAXLIST/Radial.zip

and here is an image demonstrating the issue:

http://www.abstrakt.vade.info/ILOVETHEMAXLIST/radial.jpg

Id love to know how this is implemented elsewhere, but I havent found 

much info. I think its time I sprung for an image processing book :(

Ive implemented polar to cartesian and cartesian to polar shaders  
(these work like the photoshop filters, c74s is a little odd on my  
system) that allow doing some nice blurring radially. However, I am  
getting a vertical line in my output. I know this is introduced by  
the polar conversion, and so I ask, is there another mapping or polar  
type mode that keeps the edges of the cartesian image in line with  
the edges of the polar image, so it does not introduce any odd  
artifacts? It makes a bit more sense when you see an image.

Here are the shaders and a patch (these will be part of a release, so  
feel free to use them),

Id love to know how this is implemented elsewhere, but I havent found  
much info. I think its time I sprung for an image processing book :(

glsl-radial-and-zoom-blurs-working-but-with-hiccup

> Ive implemented polar to cartesian and cartesian to polar shaders 

> (these work like the photoshop filters, c74s is a little odd on my 

> system) that allow doing some nice blurring radially. However, I am 

> getting a vertical line in my output. I know this is introduced by 

> the polar conversion, and so I ask, is there another mapping or 

> polar type mode that keeps the edges of the cartesian image in line 

> with the edges of the polar image, so it does not introduce any odd 

> artifacts? It makes a bit more sense when you see an image.

This is basically unavoidable the way you're approaching it. It 

sounds like you might want to make something like the "twirl" 

shader, or another idea would be to mirror the polar image so that 

it tiles by mirroring back on itself, though this is probably not 

what you're looking to do. A final idea might be to consider polar 

phase differences (i.e. consider the polar mapping a phase increment 

by which to displace, rather than absolute coordinates). For the 

point where the line happens (~2pi->~0) consider it to be a small 

number (positive modulo of 2pi like you would have with a sine or 

cosine function). The amplify the relative phase to grab texture 

coordinates, rather than the absolute phase. Although, this might 

work out to be roughly exactly what the twirl shader does. Haven't 

> Ive implemented polar to cartesian and cartesian to polar shaders  
> (these work like the photoshop filters, c74s is a little odd on my  
> system) that allow doing some nice blurring radially. However, I am  
> getting a vertical line in my output. I know this is introduced by  
> the polar conversion, and so I ask, is there another mapping or  
> polar type mode that keeps the edges of the cartesian image in line  
> with the edges of the polar image, so it does not introduce any odd  
> artifacts? It makes a bit more sense when you see an image.

This is basically unavoidable the way you're approaching it. It  
sounds like you might want to make something like the "twirl"  
shader,  or another idea would be to mirror the polar image so that  
it tiles by mirroring back on itself, though this is probably not  
what you're looking to do. A final idea might be to consider polar  
phase differences (i.e. consider the polar mapping a phase increment  
by which to displace, rather than absolute coordinates). For the  
point where the line happens (~2pi->~0) consider it to be a small  
number (positive modulo of 2pi like you would have with a sine or  
cosine function). The amplify the relative phase to grab texture  
coordinates, rather than the absolute phase. Although, this might  
work out to be roughly exactly what the twirl shader does. Haven't  
looked at it in detail.

Thanks Joshua for your feedback, its appreciated as always.

My only concern with doing things like the twirl shader is that they 

are self contained. I suppose that isnt an issue if I just wanted a 

radial blur or zoom blur, but the nice (or, perhaps interesting) 

thing about having discreet steps as shaders is that you can nest 

operations/steps very easily - like putting something besides the 

guassian blurs in between the car to pol pol to car, etc, result in 

different effects, but I suppose I just have to write separate 

I found a nice Nvidia HLSL example I am going to eventually port to 

GLSL for the blurs. My concern is the 'feel'. I really like the 

whispyness of the current solution. Oh well.

Anyway, thanks again, ill keep poking at it in my (not so free) free 

On Aug 26, 2007, at 11:35 PM, Joshua Kit Clayton wrote:

> On Aug 26, 2007, at 7:42 PM, vade wrote:

>> Ive implemented polar to cartesian and cartesian to polar shaders 

>> (these work like the photoshop filters, c74s is a little odd on my 

>> system) that allow doing some nice blurring radially. However, I 

>> am getting a vertical line in my output. I know this is introduced 

>> by the polar conversion, and so I ask, is there another mapping or 

>> polar type mode that keeps the edges of the cartesian image in 

>> line with the edges of the polar image, so it does not introduce 

>> any odd artifacts? It makes a bit more sense when you see an image.

> This is basically unavoidable the way you're approaching it. It 

> sounds like you might want to make something like the "twirl" 

> shader, or another idea would be to mirror the polar image so that 

> it tiles by mirroring back on itself, though this is probably not 

> what you're looking to do. A final idea might be to consider polar 

> phase differences (i.e. consider the polar mapping a phase 

> increment by which to displace, rather than absolute coordinates). 

> For the point where the line happens (~2pi->~0) consider it to be a 

> small number (positive modulo of 2pi like you would have with a 

> sine or cosine function). The amplify the relative phase to grab 

> texture coordinates, rather than the absolute phase. Although, this 

> might work out to be roughly exactly what the twirl shader does. 

My only concern with doing things like the twirl shader is that they  
are self contained. I suppose that isnt an issue if I just wanted a  
radial blur or zoom blur, but the nice (or, perhaps interesting)  
thing about having discreet steps as shaders is that you can nest  
operations/steps very easily - like putting something besides the  
guassian blurs in between the car to pol pol to car, etc, result in  
different effects, but I suppose I just have to write separate  
shaders then.

I found a nice Nvidia HLSL example I am going to eventually port to  
GLSL for the blurs. My concern is the 'feel'. I really like the  
whispyness of the current solution. Oh well.

Anyway, thanks again, ill keep poking at it in my (not so free) free  
time.

>
> On Aug 26, 2007, at 7:42 PM, vade wrote:
>
>> Ive implemented polar to cartesian and cartesian to polar shaders  
>> (these work like the photoshop filters, c74s is a little odd on my  
>> system) that allow doing some nice blurring radially. However, I  
>> am getting a vertical line in my output. I know this is introduced  
>> by the polar conversion, and so I ask, is there another mapping or  
>> polar type mode that keeps the edges of the cartesian image in  
>> line with the edges of the polar image, so it does not introduce  
>> any odd artifacts? It makes a bit more sense when you see an image.
>
>
> This is basically unavoidable the way you're approaching it. It  
> sounds like you might want to make something like the "twirl"  
> shader,  or another idea would be to mirror the polar image so that  
> it tiles by mirroring back on itself, though this is probably not  
> what you're looking to do. A final idea might be to consider polar  
> phase differences (i.e. consider the polar mapping a phase  
> increment by which to displace, rather than absolute coordinates).  
> For the point where the line happens (~2pi->~0) consider it to be a  
> small number (positive modulo of 2pi like you would have with a  
> sine or cosine function). The amplify the relative phase to grab  
> texture coordinates, rather than the absolute phase. Although, this  
> might work out to be roughly exactly what the twirl shader does.  
> Haven't looked at it in detail.
>
> -Joshua

GLSL Radial and Zoom blurs - working but with hiccup.