topic Erode3x3/Dilate3x3 inplace - Implementation for other platforms in Intel® Integrated Performance Primitives

Erode3x3/Dilate3x3 inplace - Implementation for other platforms

mweissbach — Tue, 11 Apr 2006 16:29:46 GMT

Hello everybody

I want to implement functions for dilating and eroding _inplace_ an image with an 3x3 filterkernel. I used for computing in Intel machines the IPP 4.0 . Now I want to implement it for other architectures by hand in Ansi C. The problem is, that it seems to be impossible to do that inplace - without further dynamical allocation of memory beside the image itself.

Now I want to know, if someone had the same problem or provides a solution for this. How is it solved in the IPP internally?

Thanks a lot for your support

M.W.

Re: Erode3x3/Dilate3x3 inplace - Implementation for other platf

Intel_C_Intel — Tue, 11 Apr 2006 22:26:06 GMT

Erosion/dilation for the rectangular kernel is separable. So you can do it as the separable convolution

Alexander

Re: Erode3x3/Dilate3x3 inplace - Implementation for other platf

mweissbach — Wed, 12 Apr 2006 13:34:38 GMT

akibkal1 wrote:
Erosion/dilation for the rectangular kernel is separable. So you can do it as the separable convolution
Alexander

Thanks a lot for you immediate answer. Since there is no filterkernel given in the IPP Manual/Reference (Version 4.1), it is hard to find the correct one, so that the results of the computing with IPP and Ansi-C without using IPP are equal. Is there a possibility to publish the filterkernels of the functions mentioned above?

Thanks

M.W.

Re: Erode3x3/Dilate3x3 inplace - Implementation for other platf

Intel_C_Intel — Wed, 12 Apr 2006 14:04:47 GMT

Erode3x3/Dilate3x3 use gray-scale image and binary kernel with all 1s. Ancor is in the center. For arbitrary binary kernel you could use general morphology function.

Alexander