What is the best approach towards finding a grid?

[Avital Steinberg]

Hi,

We are analyzing images where we need to quantify the size of "white blobs"
arranged in a grid format. The grid is organized in equal spacing in the x
and y directions. Some blobs might be missing, and the blobs may be on a
slightly slanted line (i.e., the grid may be rotated by up to 10-15
degrees).

My problem is to locate the grid. The simple way to do this is the
following:

for each possible Xstart
---for each possible Ystart
-----for each possible angle,
--------- sum the whiteness at each position of the grid
At the end, keep the X/Y/angle that maximize whiteness.

The problem is that the grid contains 1536 points and so each iteration
takes a lot of time and there are ~16000 of them (40 Xstart * 40 Ystart *
10 angles). I was thus wondering if one could speedup the "whiteness
measurement" by summing simultaneaously all the pixels corresponding to the
grid position? At the moment, we have 1536 ROIs and this is way too slow.

Is there a more efficient way to approach this problem?

Thank you,
Avital

--
ImageJ mailing list: http://imagej.nih.gov/ij/list.html

[Michael Schmid]

Hi Avital,

what about FFT, select the first-order maxima and fill them with 255, run Inverse FFT and "Find Maxima"?
Depending on the image data, you might want to do some preprocessing to suppress the background.

Michael
________________________________________________________________
On Nov 13, 2014, at 17:46, Avital Steinberg wrote:

> Hi,
>
> We are analyzing images where we need to quantify the size of "white blobs"
> arranged in a grid format. The grid is organized in equal spacing in the x
> and y directions. Some blobs might be missing, and the blobs may be on a
> slightly slanted line (i.e., the grid may be rotated by up to 10-15
> degrees).
>
> My problem is to locate the grid. The simple way to do this is the
> following:
>
> for each possible Xstart
> ---for each possible Ystart
> -----for each possible angle,
> --------- sum the whiteness at each position of the grid
> At the end, keep the X/Y/angle that maximize whiteness.
>
> The problem is that the grid contains 1536 points and so each iteration
> takes a lot of time and there are ~16000 of them (40 Xstart * 40 Ystart *
> 10 angles). I was thus wondering if one could speedup the "whiteness
> measurement" by summing simultaneaously all the pixels corresponding to the
> grid position? At the moment, we have 1536 ROIs and this is way too slow.
>
> Is there a more efficient way to approach this problem?
>
> Thank you,
> Avital
>
> --
> ImageJ mailing list: http://imagej.nih.gov/ij/list.html

--
ImageJ mailing list: http://imagej.nih.gov/ij/list.html

[Cammer, Michael]

Maybe do an fft, blur, find maxima, filter out all but the maxima, deconvolve, threshold, analyze particles.

===========================================================================
Michael Cammer, Microscopy Core & Skirball Institute, NYU Langone Medical Center
Cell:  914-309-3270   note that we do not receive messages left at 212-263-3208
http://ocs.med.nyu.edu/microscopy & http://microscopynotes.com/


-----Original Message-----
From: ImageJ Interest Group [mailto:[hidden email]] On Behalf Of Avital Steinberg
Sent: Thursday, November 13, 2014 11:46 AM
To: [hidden email]
Subject: What is the best approach towards finding a grid?

Hi,

We are analyzing images where we need to quantify the size of "white blobs"
arranged in a grid format. The grid is organized in equal spacing in the x and y directions. Some blobs might be missing, and the blobs may be on a slightly slanted line (i.e., the grid may be rotated by up to 10-15 degrees).

My problem is to locate the grid. The simple way to do this is the
following:

for each possible Xstart
---for each possible Ystart
-----for each possible angle,
--------- sum the whiteness at each position of the grid At the end, keep the X/Y/angle that maximize whiteness.

The problem is that the grid contains 1536 points and so each iteration takes a lot of time and there are ~16000 of them (40 Xstart * 40 Ystart *
10 angles). I was thus wondering if one could speedup the "whiteness measurement" by summing simultaneaously all the pixels corresponding to the grid position? At the moment, we have 1536 ROIs and this is way too slow.

Is there a more efficient way to approach this problem?

Thank you,
Avital

--
ImageJ mailing list: http://imagej.nih.gov/ij/list.html

--
ImageJ mailing list: http://imagej.nih.gov/ij/list.html

[Avital Steinberg]

Thank you for your suggestions - I will try your ideas out.

Have a nice weekend,
Avital

On Thu, Nov 13, 2014 at 7:14 PM, Cammer, Michael <[hidden email]
> wrote:

> Maybe do an fft, blur, find maxima, filter out all but the maxima,
> deconvolve, threshold, analyze particles.
>
> ===========================================================================
> Michael Cammer, Microscopy Core & Skirball Institute, NYU Langone Medical
> Center
> Cell:  914-309-3270   note that we do not receive messages left at
> 212-263-3208
> http://ocs.med.nyu.edu/microscopy & http://microscopynotes.com/
>
>
> -----Original Message-----
> From: ImageJ Interest Group [mailto:[hidden email]] On Behalf Of
> Avital Steinberg
> Sent: Thursday, November 13, 2014 11:46 AM
> To: [hidden email]
> Subject: What is the best approach towards finding a grid?
>
> Hi,
>
> We are analyzing images where we need to quantify the size of "white blobs"
> arranged in a grid format. The grid is organized in equal spacing in the x
> and y directions. Some blobs might be missing, and the blobs may be on a
> slightly slanted line (i.e., the grid may be rotated by up to 10-15
> degrees).
>
> My problem is to locate the grid. The simple way to do this is the
> following:
>
> for each possible Xstart
> ---for each possible Ystart
> -----for each possible angle,
> --------- sum the whiteness at each position of the grid At the end, keep
> the X/Y/angle that maximize whiteness.
>
> The problem is that the grid contains 1536 points and so each iteration
> takes a lot of time and there are ~16000 of them (40 Xstart * 40 Ystart *
> 10 angles). I was thus wondering if one could speedup the "whiteness
> measurement" by summing simultaneaously all the pixels corresponding to the
> grid position? At the moment, we have 1536 ROIs and this is way too slow.
>
> Is there a more efficient way to approach this problem?
>
> Thank you,
> Avital
>
> --
> ImageJ mailing list: http://imagej.nih.gov/ij/list.html
>
> --
> ImageJ mailing list: http://imagej.nih.gov/ij/list.html
>

--
ImageJ mailing list: http://imagej.nih.gov/ij/list.html