[Date Index] [Thread Index] [Author Index]

Re: Circular neighborhood for ImageApply

• To: mathgroup at smc.vnet.net
• Subject: [mg115064] Re: Circular neighborhood for ImageApply
• From: Shadi Ashnai <shadi at wolfram.com>
• Date: Thu, 30 Dec 2010 04:09:40 -0500 (EST)

Hi Adrian, 

In ImageFilter you cannot specify a non-rectangular neighborhood yet. The following code would only count the zero pixels inside the disk and divide by the size of the disk: 

disksize = Total[DiskMatrix[r], 2]; 
ImageFilter[Count[DiskMatrix[r] - #, 1., 2]/disksize &, ColorConvert[i, "Grayscale"], r] 



Another suggestion would be to remove disks smaller than your radius by doing: 
Closing[i, DiskMatrix[6]] 

If you got Mathematica 8, you can simply compute the center of each foreground object (remaining disks) by either of the following: 

1) MaxDetect[ DistanceTransform[ image ] ] (* computes the ultimate erosion and returns an image consisting of the centers *) 

2) ComponentMeasurements[ image, "Centroid" ] (* returns the coordinate of the centers *) 


Best, 

Shadi Ashnai 
Wolfram Research Inc. 






From: "Adrian Boyko" <adrianboyko at gmail.com> 
To: mathgroup at smc.vnet.net 
Sent: Tuesday, December 28, 2010 5:52:36 AM 
Subject: [mg115064] [mg115022] Circular neighborhood for ImageApply 

Usual disclaimer: I'm a casual home user of Mathematica, so please 
forgive what might seem like silly questions. 

I'm using ImageFilter like this: 

ImageFilter[1.0*Count[Flatten[#], 0.]/Length[Flatten[#]] &, 
ColorConvert[binaryImage, "Grayscale"], r] 

I.e. I'm trying to replace each pixel with the ratio of 0 pixels and 1 
pixels in its square neighborhood. I'm doing this because I'm trying to 
find the center of disks (of an expected radius) in a scanned image, and 
this works reasonably well. But, since I'm actually looking for disks I 
think it would be better to find the ratio in the pixel's _disk_ (not 
square) neighborhood. It seems like there ought to be a simple way to 
work DiskMatrix[3] into my expression, and extra processing time isn't 
an issue. 

Any ideas? Is there some operator I can apply to # and DiskMatrix[3] 
which would mask out the neighborhood values outside the disk? I think 
ImageMultiply ought to work like an ImageAnd if all my pixel values are 
0 or 1, but I haven't had any luck with it. 

On a tangent, I have another filter that ought to be looking for the 
ratio in a ring neighborhood between r1 and r2. Maybe the answer to my 
first question will also tell me how to combine DiskMatrix[r1] and 
DiskMatrix[r2] to get RingMatrix[r1,r2]. 

Any help will be greatly appreciated. 

Thanks, 
Adrian 

P.S. I'm considering the upgrade to Mathematica 8. Since I have a 
CUDA-compatible graphics card, is it safe to assume that version 8 will 
_automatically_ utilize it when executing these sorts of operations? Or 
would I need to code differently to take advantage of it?