[Insight-users] ConvolutionImageFilter only works with scalar images?

David Doria daviddoria at gmail.com
Sat Dec 4 16:15:19 EST 2010


On Sat, Dec 4, 2010 at 4:06 PM, Gelas, Arnaud Joel Florent
<Arnaud_Gelas at hms.harvard.edu> wrote:
> Hi David,
>
> The problem with this filter is that the kernel image is of the same type as the input image.
> I guess this filter needs an additional template argument for the kernel image (a scalar image), like this the input image could theoretically be of any type (scalar, vector, tensor) ?
>
> Arnaud

If the KernelType was set to a scalar, what you suggest would do the
trick (there would be a scalar multiplication of each Image pixel).
However, if you want to convolve an RGB image with another RGB image,
this still doesn't make sense, right? The channels would have to be
convolved separately (because RGBPixel * RGBPixel should be
interpreted as Rout = R1 * R2, Gout = G1*G2, Bout = B1*B2, and that
idea isn't the behavior of the itk::CovariantVector '*' operator).

David


More information about the Insight-users mailing list