Round object detection

asked 2013-03-09 18:00:54 -0600

This post is a wiki. Anyone with karma >50 is welcome to improve it.

image description Hello,

I am new to opencv and I would like some hints. I am doing clinical research and I need to detect sperm cells by analyzing images. The pictures are in grayscale, need to cleaned out of impurities and contrast must be increased. The magnification is low, so the cells are barely visible to the human eye. I have tried with templates, but doesn't quite work, because the position/orientation varies. What is the best approach for filtering and detecting an elliptical object with a tail? Here is a link with an image: http://www.profilekiss.com/picture/code-1/sperm.html or http://media.beta.photobucket.com/user/DSCDBIO/media/frogspermsmear100X.jpg.html?filters[term]=sperm%20100X&filters[primary]=images&filters[secondary]=videos&sort=1&o=0 image description Tx

Comments

template matching won't work ( as you found out before ) because it only compares a rect of pixels, and your sperms there can have any rotation or even bend to the one or other side.

what about adding a (small) demo pic here, so people can try out ideas ?

berak ( 2013-03-10 17:05:47 -0600 )edit

add a comment

answered 2013-03-11 04:55:55 -0600

rics

1093 ●1 ●14 ●21

You may try to use Hough circle transform.

This image processing page also mentions the Hough circle transfrom as a possibility on cell detection.

Here is a short overview on approaching a problem similar to yours using MatLab.

Here is an article from last year on blood cell detection using Hough circle. The authors report that the automatic counting was quite accurate.

Here is an even more fresh study on white blood cells. The introduction enumerates some extensions of Hough circle transform and even other approaches like genetic algorithms.

edit flag offensive delete link

Comments

Nice guidance to the problem!

StevenPuttemans ( 2013-03-11 10:48:42 -0600 )edit

I don't think this will work since sperm cells are not round, are elliptical, and main feature to rule out false positives is the tail. The picture below is poor in quality, but can give you an ideea:

http://media.beta.photobucket.com/user/DSCDBIO/media/frogspermsmear100X.jpg.html?filters[term]=sperm%20100X&filters[primary]=images&filters[secondary]=videos&sort=1&o=0

sf102 ( 2013-03-18 06:26:06 -0600 )edit

Here is a paper on how to detect ellipses with Randomized Hough Transform: http://www.saminverso.com/res/vision/EllipseDetection.pdf

rics ( 2013-03-18 07:11:30 -0600 )edit

add a comment

Round object detection

Comments

1 answer

Comments

Links

Question Tools

Stats

Related questions

Round object detection edit

Comments

1 answer

Comments

Links

Question Tools

Stats

Related questions

Round object detection