Registration error for certain image sizes

[swift505]

Hi,

I'm getting errors during registration for certain values of image size. I therefore ran some tests, trying to register the same movie after cropping to different sizes.

My movie is 1024x1024 pixels x 881 frames. This works fine, and also registers if I crop to 767x768 or 512x512 pixels. However, 1023x1023 and 767x767 cause crashes as per below:

---

ERROR with 1023x1023 (x881 frames) and with 767x767 (x881 frames)::

running rigid registration
Error using .*
Array dimensions must match for binary
array op.

Error in BiDiPhaseOffsets (line 18)
cc = ifft(d1 .* d2,[],2);

Error in reg2P (line 44)
BiDiPhase = BiDiPhaseOffsets(IMG);

---

[marius10p]

Seems like this has to do with BiDiPhaseOffsets. Any ideas @carsen-stringer ?

[carsen-stringer]

ah yes sorry this was only working for frames with even pixels! try now

[swift505]

Hi @carsen-stringer
I now no longer get the error messages, but the registered TIFFs show a strange 'scrolling' artefact where the image appears to translate vertically by approximately 1 row per frame. This only occurs for odd numbers of pixels.
768x768 works fine -0.2 0 0 0.1
1024x1024 works fine -0.2 0 0 0
767x767 scrolling effect -0.2 0 0 0.1
1023x1023 scrolling effect -0.2 0 0 0

[carsen-stringer]

Are they giving different bidiphase offsets depending on the number of pixels? Or is the bidiphase offset the same for both?

[swift505]

Hi @carsen-stringer,
In all cases bidir offset = 0.

• Isaac

[carsen-stringer]

that's strange, now it should avoid any pixel shifts if bidiphase=0. let me know if it works.

[swift505]

Hi @carsen-stringer ,

I'm afraid the 'scrolling' problem still seems to persist. Also, if this helps diagnose the problem, as well as translating vertically 1 row per frame, the registered movie also appears to have jumped horizontally (by about 80 pixels). As usual, this did not happen with a 768x768 version of the same data.

My settings in master_file for registration are:
% registration options
ops0.doRegistration = 1; % skip (0) if data is already registered
ops0.showTargetRegistration = 1; % shows the image targets for all planes to be registered
ops0.PhaseCorrelation = 1; % set to 0 for non-whitened cross-correlation
ops0.SubPixel = Inf; % 2 is alignment by 0.5 pixel, Inf is the exact number from phase correlation
ops0.NimgFirstRegistration = 100; % was 500, number of images to include in the first registration pass
ops0.nimgbegend = 100; % was 250, frames to average at beginning and end of blocks
ops0.nonrigid = 0; % IHB added

The dataset I just tested that failed is 767x767 pixels, x 881 frames, 16-bit TIF.
The programme reported:

---

running rigid registration
bi-directional scanning offset = 0 pixels
Set 1, tiff 1 done in time 50.96
-0.2000 0 0 0.1000

---

• Isaac

[carsen-stringer]

What are your ops.Ly and ops.Lx? Are they 767?

[swift505]

@carsen-stringer
ops1.Lx and ops1.Ly are 767, yes.
(The command I'm running is ops1 = reg2P(ops);
ops itself doesn't seem to have an Lx/Ly field so I assume these are added by the registration function.)

[carsen-stringer]

Can you host this tiff on dropbox or google drive and share it with me (carsen.stringer@gmail.com)? Sorry I don't know where the script is failing

[carsen-stringer]

the error could not be replicated