[3dem] Edge bitmap curvature analysis

[Ludtke, Steven J.]

Hi George,
It's no problem, and a lot of image processing people do read this list, but it is mainly focused on CryoEM and CryoET. Both of these produce very noisy data, on which edge detection algorithms often don't work well (or at all).  You may find some people in this list with comments on your specific question, particularly from those doing freeze-substitution and other higher contrast lower resolution methods.

For myself (EMAN2), for discrete edges like binary masks, we sometimes use standard methods such as erosion or dilation for expansion/contraction. However, for a smoother edge following effect, our typical approach is to binarize the mask (if it isn't already) apply an isotropic low-pass filter, then threshold the filter at a lower level. By adjusting the thresholding level and filter shape a range of different 'curve following' effects can be achieved. Certainly there are many different approaches which can be used. By and large we have eschewed the sort of processing where shapes are explicitly identified, due to the noise levels.

There is a tool in EMAN2 called e2filtertool.py in which you can build chains of image processing operations and interactively adjust the parameters (both 2-D and 3-D data). You can use this to experiment with any of the methods we have implemented (~230 algorithms at last count, ranging from very simple, to very complicated).  We also have a set of programs for deep-learning based segmentation, which can often do some surprising things.

cheers


--------------------------------------------------------------------------------------
Steven Ludtke, Ph.D. <sludtke at bcm.edu<mailto:sludtke at bcm.edu>>                      Baylor College of Medicine
Charles C. Bell Jr., Professor of Structural Biology
Dept. of Biochemistry and Molecular Biology                      (www.bcm.edu/biochem<http://www.bcm.edu/biochem>)
Academic Director, CryoEM Core                                        (cryoem.bcm.edu<http://cryoem.bcm.edu>)
Co-Director CIBR Center                                    (www.bcm.edu/research/cibr<http://www.bcm.edu/research/cibr>)



On Jul 29, 2019, at 12:59 PM, Glidden, George (NIH/NIAID) [F] <george.glidden at nih.gov<mailto:george.glidden at nih.gov>> wrote:

I should have been more specific – the algorithm is specifically for use in the segmentation of EM data, wherein this algorithm would be one step in preprocessing, with inpainted edges allowing for better extractions of regions of interest.
Apologies if this was the wrong list to post this question to.

From: Ludtke, Steven J. <sludtke at bcm.edu<mailto:sludtke at bcm.edu>>
Sent: Monday, July 29, 2019 11:55 AM
To: Glidden, George (NIH/NIAID) [F] <george.glidden at nih.gov<mailto:george.glidden at nih.gov>>
Subject: Re: [3dem] Edge bitmap curvature analysis

Hi George, not quite clear how this is cryoem-related? What sort of images are you analyzing and what exactly arw you trying to accomplish? As a general question seems more appropriate for a CS/computer vision list.
Sent from my iPhone

On Jul 29, 2019, at 12:19 PM, Glidden, George (NIH/NIAID) [F] <george.glidden at nih.gov<mailto:george.glidden at nih.gov>> wrote:
***CAUTION:*** This email is not from a BCM Source. Only click links or open attachments you know are safe.
________________________________
Hi all,

First time posting to this list and unsure of how long / how much detail to include in the emails, so I’ll keep this as brief as possible:
I’ve been doing work into edge inpainting approaches using information from the curvature of the existing edges to recreate the occluded or otherwise missing points.
One method I have recently developed divides edge strips – continuous lists of adjacent edge pixels without branches – into curves, split on inflection points calculated from the distance between each point and an approximated center of the strip; i.e., the inflection points of the curve of a vector-valued function. In other words, the strips are split into regions of uniform concavity - concavity relative to the strip, not to a cartesian coordinate plane.
(My definition of edge strips is based on the definition given in this publication<https://urldefense.proofpoint.com/v2/url?u=https-3A__www.researchgate.net_publication_261999152-5FA-5FFast-5Fand-5FRobust-5FEllipse-2DDetection-5FMethod-5FBased-5Fon-5FSorted-5FMerging&d=DwMFAg&c=ZQs-KZ8oxEw0p81sqgiaRA&r=Dk5VoQQ-wINYVssLMZihyC5Dj_sWYKxCyKz9E4Lp3gc&m=NCJUJybdIpPQDddFIr8a-hCtUiboMu5alfSyxooTDTE&s=nqXemHJADlGIKg_QLDshiMl-6qMFvjv2qLghfZ1vCgc&e=>, and the my scripts process the results of modified steps 1 and 2 in 2.1 Line Segment Extraction.)
Has anyone previously encountered and solved a problem like this, or otherwise, are there suggestions for better, more mathematically rigorous approaches to this problem (ones that don’t rely on as much approximation)?
I’m required to list all of my github repositories privately, but I will happily individually provide access to anyone interested in this problem.

Thank you!

George
_______________________________________________
3dem mailing list
3dem at ncmir.ucsd.edu<mailto:3dem at ncmir.ucsd.edu>
https://urldefense.proofpoint.com/v2/url?u=https-3A__mail.ncmir.ucsd.edu_mailman_listinfo_3dem&d=DwICAg&c=ZQs-KZ8oxEw0p81sqgiaRA&r=Dk5VoQQ-wINYVssLMZihyC5Dj_sWYKxCyKz9E4Lp3gc&m=NCJUJybdIpPQDddFIr8a-hCtUiboMu5alfSyxooTDTE&s=lBdaTBxvpnjz77DxykFEC3FBZCpU056WfZdAwJ5DzcE&e=

-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://mail.ncmir.ucsd.edu/pipermail/3dem/attachments/20190729/7e8af529/attachment-0001.html>