What is the thing with "Test Block matching parameters"?

[Christin Wittfoth]

Hi everyone,

we try to find out how to do an elastic alignment but we despair of
testing block matching parameters in TrakEM2. We are not sure how to
interpret the results and to decide if we have found the right
parameters. We attached one snapshot as example:
- Is the size of the black area important? (smaller <-> bigger)
- What do the colours mean?
- How important is the blackness of the colours? (Compare the middle row
and column which show bright colours but in complementary ones of the
colour range)
- We have read the parameters which are defined in pixels depend on the
image size. But in which range can we test the parameters to get a
successful alignment?
- What does the layer scale mean and is depending on?
- Can we use information out of the log? (We noticed the log name a
"maximum local displacement", for example.)

The size of the images was 3000 x 2250 px.

To perform a pre-alignment we used as parameters:

steps per scale octave: 3
minimum image size: 256
maximum image size: 1024
feature extraction threads: 3
maximal alignment error: 225
minimal inlier ratio: 0.05
minimal numbers of inliers: 15
expected transformation: Affine
test multiple hypotheses (max. 3 layers, give up after 3 failures)
desired transformation: Rigid
maximal iterations: 1000
plateauwidth: 200

We exported the flat images by using a scale of 50 % and tested the
block matching parameters:

layer scale: 0.33
search radius: 115
block radius: 145
resolution: 30
approximate local transformation: Rigid
local region sigma: 65
maximal local displacement - absolute: 135
- relative: 35


We are grateful for every help!

Greetings,
Christin & Tina

------------------------------------------------------------

M.Sc. Christin Wittfoth

Cytologie und Evolutionsbiologie
Universität Greifswald
Zoologisches Institut und Museum
Soldmannstrasse 23
Haus 6.1
D - 17487 Greifswald
Tel.: [phone]
Fax: [phone]


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

[Saalfeld, Stephan]

Hi Christin,

blind shot.  Try this:

To perform a pre-alignment we used as parameters:

steps per scale octave: 3
minimum image size: 256
maximum image size: 1024
feature extraction threads: 3
maximal alignment error: 50
minimal inlier ratio: 0
minimal numbers of inliers: 5
expected transformation: Rigid
test multiple hypotheses: yes
test max. 3 layers, give up after 3 failures
desired transformation: Affine
regularize: yes -> Rigid, lambda 0.1
maximal iterations: 1000
plateauwidth: 200

Why?  Section 3 appears to be unaligned to the rest of the stack.  If
that could be fixed, search for elastic distortion becomes a lot
simpler and faster.  Then:

Test Blockmatching:

layer scale: 0.25
search radius: 50
block radius: 75
resolution: 30
smoothness filter (yes)
approximate local transformation: Rigid
local region sigma: 150
maximal local displacement - absolute: 20
- relative: 3

yet, this may be wrong, depending on what happened before.  Did you
read

http://imagej.net/Test_Block_Matching_Parameters

?  That Wiki-page explains what the colors and blackness mean.

Best,
Stephan





On Wed, 2016-11-02 at 09:43 +0100, Christin Wittfoth wrote:

> Hi everyone,
>
> we try to find out how to do an elastic alignment but we despair of 
> testing block matching parameters in TrakEM2. We are not sure how to 
> interpret the results and to decide if we have found the right 
> parameters. We attached one snapshot as example:
> - Is the size of the black area important? (smaller <-> bigger)
> - What do the colours mean?
> - How important is the blackness of the colours? (Compare the middle
> row 
> and column which show bright colours but in complementary ones of
> the 
> colour range)
> - We have read the parameters which are defined in pixels depend on
> the 
> image size. But in which range can we test the parameters to get a 
> successful alignment?
> - What does the layer scale mean and is depending on?
> - Can we use information out of the log? (We noticed the log name a 
> "maximum local displacement", for example.)
>
> The size of the images was 3000 x 2250 px.
>
> To perform a pre-alignment we used as parameters:
>
> steps per scale octave: 3
> minimum image size: 256
> maximum image size: 1024
> feature extraction threads: 3
> maximal alignment error: 225
> minimal inlier ratio: 0.05
> minimal numbers of inliers: 15
> expected transformation: Affine
> test multiple hypotheses (max. 3 layers, give up after 3 failures)
> desired transformation: Rigid
> maximal iterations: 1000
> plateauwidth: 200
>
> We exported the flat images by using a scale of 50 % and tested the 
> block matching parameters:
>
> layer scale: 0.33
> search radius: 115
> block radius: 145
> resolution: 30
> approximate local transformation: Rigid
> local region sigma: 65
> maximal local displacement - absolute: 135
> - relative: 35
>
>
> We are grateful for every help!
>
> Greetings,
> Christin & Tina
>
> ------------------------------------------------------------
>
> M.Sc. Christin Wittfoth
>
> Cytologie und Evolutionsbiologie
> Universität Greifswald
> Zoologisches Institut und Museum
> Soldmannstrasse 23
> Haus 6.1
> D - 17487 Greifswald
> Tel.: [phone]
> Fax: [phone]
>
>
> --
> ImageJ mailing list: http://imagej.nih.gov/ij/list.html

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

[Christin Wittfoth]

Hi Stephan,

thank you very much! I will try the parameters you recommended.

I thought I read this page but I could not remember there are advices
for changing parameters: so thanks again. I will read it more carefully!

Best,

Christin


Am 03.11.2016 um 19:20 schrieb Saalfeld, Stephan:

> Hi Christin,
>
> blind shot.  Try this:
>
> To perform a pre-alignment we used as parameters:
>
> steps per scale octave: 3
> minimum image size: 256
> maximum image size: 1024
> feature extraction threads: 3
> maximal alignment error: 50
> minimal inlier ratio: 0
> minimal numbers of inliers: 5
> expected transformation: Rigid
> test multiple hypotheses: yes
> test max. 3 layers, give up after 3 failures
> desired transformation: Affine
> regularize: yes -> Rigid, lambda 0.1
> maximal iterations: 1000
> plateauwidth: 200
>
> Why?  Section 3 appears to be unaligned to the rest of the stack.  If
> that could be fixed, search for elastic distortion becomes a lot
> simpler and faster.  Then:
>
> Test Blockmatching:
>
> layer scale: 0.25
> search radius: 50
> block radius: 75
> resolution: 30
> smoothness filter (yes)
> approximate local transformation: Rigid
> local region sigma: 150
> maximal local displacement - absolute: 20
> - relative: 3
>
> yet, this may be wrong, depending on what happened before.  Did you
> read
>
> http://imagej.net/Test_Block_Matching_Parameters
>
> ?  That Wiki-page explains what the colors and blackness mean.
>
> Best,
> Stephan
>
>
>
>
>
> On Wed, 2016-11-02 at 09:43 +0100, Christin Wittfoth wrote:
>> Hi everyone,
>>
>> we try to find out how to do an elastic alignment but we despair of
>> testing block matching parameters in TrakEM2. We are not sure how to
>> interpret the results and to decide if we have found the right
>> parameters. We attached one snapshot as example:
>> - Is the size of the black area important? (smaller <-> bigger)
>> - What do the colours mean?
>> - How important is the blackness of the colours? (Compare the middle
>> row
>> and column which show bright colours but in complementary ones of
>> the
>> colour range)
>> - We have read the parameters which are defined in pixels depend on
>> the
>> image size. But in which range can we test the parameters to get a
>> successful alignment?
>> - What does the layer scale mean and is depending on?
>> - Can we use information out of the log? (We noticed the log name a
>> "maximum local displacement", for example.)
>>
>> The size of the images was 3000 x 2250 px.
>>
>> To perform a pre-alignment we used as parameters:
>>
>> steps per scale octave: 3
>> minimum image size: 256
>> maximum image size: 1024
>> feature extraction threads: 3
>> maximal alignment error: 225
>> minimal inlier ratio: 0.05
>> minimal numbers of inliers: 15
>> expected transformation: Affine
>> test multiple hypotheses (max. 3 layers, give up after 3 failures)
>> desired transformation: Rigid
>> maximal iterations: 1000
>> plateauwidth: 200
>>
>> We exported the flat images by using a scale of 50 % and tested the
>> block matching parameters:
>>
>> layer scale: 0.33
>> search radius: 115
>> block radius: 145
>> resolution: 30
>> approximate local transformation: Rigid
>> local region sigma: 65
>> maximal local displacement - absolute: 135
>> - relative: 35
>>
>>
>> We are grateful for every help!
>>
>> Greetings,
>> Christin & Tina
>>
>> ------------------------------------------------------------
>>
>> M.Sc. Christin Wittfoth
>>
>> Cytologie und Evolutionsbiologie
>> Universität Greifswald
>> Zoologisches Institut und Museum
>> Soldmannstrasse 23
>> Haus 6.1
>> D - 17487 Greifswald
>> Tel.: [phone]
>> Fax: [phone]
>>
>>
>> --
>> ImageJ mailing list: http://imagej.nih.gov/ij/list.html
> --
> ImageJ mailing list: http://imagej.nih.gov/ij/list.html

--
------------------------------------------------------------

M.Sc. Christin Wittfoth

Cytologie und Evolutionsbiologie
Universität Greifswald
Zoologisches Institut und Museum
Soldmannstrasse 23
Haus 6.1
D - 17487 Greifswald
Tel.: [phone]
Fax: [phone]

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