Add hessian-laplace detector #1414

warmspringwinds · 2015-03-09T00:54:05Z

Hessian Laplace blob detector efficient implementation.

ahojnnes · 2015-03-09T01:22:35Z

skimage/feature/_hessian_det_appx.pyx

@@ -151,3 +151,91 @@ def _hessian_matrix_det(cnp.double_t[:, ::1] img, double sigma):
            out[r, c] = (dxx * dyy - 0.81 * (dxy * dxy))

    return out
+
+def _hessian_matrix_det_and_laplacian(cnp.double_t[:, ::1] img, double sigma):
+    """Computes the approximate Hessian Determinant over an image


Use a single, first line as a short description.

warmspringwinds · 2015-03-10T08:25:19Z

Basically, this method is very similar to usual Hessian.
But on the coins data is shows better results. (examples are taken from source code examples)
Hessian:

Hessian-Laplace:

But it still has the problem with detecting small blobs as hessian detector:

vighneshbirodkar · 2015-03-10T10:49:25Z

skimage/feature/blob.py

+    # This points will be checked for Laplacian scal local maximum
+    list_of_peaks = []
+
+    for scale_number, current_hessian_image in enumerate(hessian_images):


Hello @warmspringwinds
I don't think you need to compute spatial maximas for each scale. From what I understand, you compute the global maximas over the image cube, and ignore the scale component which is later computed by scale space maximas for only the spacial maximas.

Hello, @vighneshbirodkar

Thank you for your review 👍

I once implemented a Harris-Laplace.
It is really similar to Hessian-Laplace, but you use Harris measure for finding local maximas
on each scale.
Here is a popular matlab's code for it:
http://www.mathworks.com/matlabcentral/fileexchange/17894-keypoint-extraction/content//keypointExtraction/kp_harrislaplace.m
You can see that they do it for each scale and then test selected points for local maxima in
Laplacian scale space.
And on the wikipedia's page it's said the same:
http://en.wikipedia.org/wiki/Blob_detection#The_Laplacian_of_Gaussian

Did I get your question right?

Yes, you did.
I think you can replace this for loop with a single call to peak_local_max with footprint as

fp = np.zeros((3,3,3)) fp[:,:,1] = 1

coveralls · 2015-03-10T19:52:43Z

Coverage increased (+0.04%) to 92.73% when pulling 1acbcb0 on warmspringwinds:hessian-laplace-blob-detector into d21b6b8 on scikit-image:master.

vighneshbirodkar · 2015-03-11T16:00:52Z

skimage/feature/peak.py

@@ -159,3 +159,73 @@ def peak_local_max(image, min_distance=10, threshold_abs=0, threshold_rel=0.1,
        nd_indices = tuple(coordinates.T)
        out[nd_indices] = True
        return out
+
+
+def get_scale_local_maximas(cube_coordinates, laplacian_cube):


This can be sped up by a Cython implementation since a majority of it's work is looping through numbers.
You can also argue that in case of _prune_blobs ( which I wrote myself )

Edit : I had adapted it from Alex Donath's code

I made some research and decided to vectorize this function instead.
I also rewrote this function in Cython, but the vectorized version works faster.
This is the related SO quetion:
http://stackoverflow.com/questions/29155745/speed-up-function-using-cython
All credits go to the author of that answer that vectorized this function.

Cython can be tricky to fully optimize, but is usually worth the effort for massive memory savings. If you wanted to share that via a Gist, we could see if the potential exists for further efficiency gains.

@JDWarner,

https://gist.github.com/warmspringwinds/1aa093ddc6f5b5ed48a6

You can see the benchmarks there.
I will be really glad if you help me.

But I think that the function is called only once in my method and won't speed up the whole script really much if we improve it further.

coveralls · 2015-03-22T00:38:04Z

Coverage increased (+0.05%) to 92.74% when pulling 826189d on warmspringwinds:hessian-laplace-blob-detector into d21b6b8 on scikit-image:master.

warmspringwinds · 2015-03-23T09:04:32Z

Dear community.
Could you, please, give me a hint on what to do next.
Because later I won't have much time and I am afraid that I won't be able to finish it.

vighneshbirodkar · 2015-03-23T09:35:37Z

@warmspringwinds An example in doc/examples would help.

vighneshbirodkar · 2015-03-23T09:37:02Z

skimage/feature/blob.py

+    The Hessian-Laplace method sometimes gives better results than the method
+    based on the determinant of Hessian. For example, on the test coin data
+    Hessian-Laplace shows better results compared to determinant of Hessian.
+    (the last method has one false positive while Hessian-Laplace doesn't)


clarify last

warmspringwinds · 2015-03-25T22:42:51Z

@vighneshbirodkar I made a thorough research and found out that Hessian-Laplace method
actually works worse than a usual method based on Determinant of Hessian.

This is because determinant of Hessian is better for scale selection as stated here:
http://link.springer.com/article/10.1007/s10851-012-0378-3/fulltext.html
You can read it in summary.

And the example with coins that I showed previously is actually is only a consequence of that fact
that Hessian-Laplace detects only bright blobs on a dark background.

And as it stated in a link that I provided above: Hessian method has
"significantly better repeatability properties under affine or perspective image transformations than the Laplacian". This is important if descriptors are later applied to these points and
matching algoritms are used to find similar points on different images
of the same scene. This blob detection method is, therefore,
is worse than Determinant of Hessian blob detection method.

All the documentation and code comments now represent my findings.

warmspringwinds · 2015-03-26T10:00:30Z

Basically this method is not so useless.
It can be, for example, used in case when the user wants to find only bright blobs
with the speed and accuracy (more or less) of Determinant of Hessian method.
Like in the example with coins (white blobs).

vighneshbirodkar · 2015-03-28T15:24:48Z

@warmspringwinds The amount of work you put in investigating is certainly appreciated. Your extra notes will certainly be appreciated.

Edit: You implementation also seems fine, why is travis complaining ?

warmspringwinds · 2015-03-29T08:12:43Z

@vighneshbirodkar there is some problem with flake8. A lot of recent pull requests have it.
I opened an issue.

warmspringwinds · 2015-03-29T14:27:39Z

@vighneshbirodkar it is a current issue that prevents travis checks to pass.
#1458

warmspringwinds · 2015-03-30T12:04:37Z

@vighneshbirodkar, now it works after rebase.

warmspringwinds · 2015-03-31T12:02:54Z

@JDWarner, may I ask you to review this PR, so that I am able to correct things?
Because later I might have not enough time.
Thank you.

ahojnnes reviewed Mar 9, 2015
View reviewed changes

vighneshbirodkar reviewed Mar 10, 2015
View reviewed changes

vighneshbirodkar reviewed Mar 11, 2015
View reviewed changes

vighneshbirodkar reviewed Mar 23, 2015
View reviewed changes

warmspringwinds added 10 commits March 30, 2015 12:10

hessian-laplace detector added

065f496

Code style corrections and failed test fix

3e711bf

forgot to remove print()

7c2826b

added missing final blank line

95962e4

small code style correction

6b2079e

peak_local_max usage added for detecting maximas on each scale

30eecd1

vectorized get_scale_local_maximas and minor documentation corrections

e6af271

examples were updated

339cc57

docs and comments were updated

bb1ad8a

documentation correction

958603b

warmspringwinds added 3 commits March 30, 2015 12:10

minor corrections

b9c3d0b

minor corrections

cf6e8c6

documentation correction

637501f

warmspringwinds force-pushed the hessian-laplace-blob-detector branch from dad76b9 to 637501f Compare March 30, 2015 10:14

soupault added the new feature label Mar 5, 2016

soupault changed the title ~~hessian-laplace detector added~~ Add hessian-laplace detector Sep 21, 2016

Base automatically changed from master to main February 18, 2021 18:22

mkcor mentioned this pull request Feb 22, 2021

2021's calendar of community management #5169

Closed

grlee77 added 🙏 Feature request and removed type: new feature labels Feb 22, 2022

mkcor mentioned this pull request Feb 28, 2022

2022's calendar of community management #6165

Closed

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Add hessian-laplace detector #1414

Add hessian-laplace detector #1414

warmspringwinds commented Mar 9, 2015

ahojnnes Mar 9, 2015

warmspringwinds commented Mar 10, 2015

vighneshbirodkar Mar 10, 2015

warmspringwinds Mar 10, 2015

vighneshbirodkar Mar 11, 2015

coveralls commented Mar 10, 2015

vighneshbirodkar Mar 11, 2015

warmspringwinds Mar 21, 2015

JDWarner Mar 21, 2015

warmspringwinds Mar 22, 2015

coveralls commented Mar 22, 2015

warmspringwinds commented Mar 23, 2015

vighneshbirodkar commented Mar 23, 2015

vighneshbirodkar Mar 23, 2015

warmspringwinds commented Mar 25, 2015

warmspringwinds commented Mar 26, 2015

vighneshbirodkar commented Mar 28, 2015

warmspringwinds commented Mar 29, 2015

warmspringwinds commented Mar 29, 2015

warmspringwinds commented Mar 30, 2015

warmspringwinds commented Mar 31, 2015

Add hessian-laplace detector #1414

Are you sure you want to change the base?

Add hessian-laplace detector #1414

Conversation

warmspringwinds commented Mar 9, 2015

Choose a reason for hiding this comment

warmspringwinds commented Mar 10, 2015

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

coveralls commented Mar 10, 2015

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

coveralls commented Mar 22, 2015

warmspringwinds commented Mar 23, 2015

vighneshbirodkar commented Mar 23, 2015

Choose a reason for hiding this comment

warmspringwinds commented Mar 25, 2015

warmspringwinds commented Mar 26, 2015

vighneshbirodkar commented Mar 28, 2015

warmspringwinds commented Mar 29, 2015

warmspringwinds commented Mar 29, 2015

warmspringwinds commented Mar 30, 2015

warmspringwinds commented Mar 31, 2015