2015/10/14

A not trivial matter

I've worked for about a month on the same image processing problem. And I feel that's a very long time. Of course, most people take about 5 years to do their PhD, but one month for a few images...


 The outline we want to detect from a fluorescent image.
 
... and then my supervisor talked to the experts in image processing on campus. And if you want to find a circle in an image, that would be easy, they said, but our shape would be "not trivial" which is great!!! What a relief that I am not stuck on adding one to one! Apparently, even entire PhD theses consist of image processing programs and there's a lot of software just for image processing.

In brief, my actual problem: There's some slightly noisy data, so if you see the image with your eyes, you think, hey, that's easy, the outline of this cute sea urchin is here, here and there, looking like the bisection of an apple. And then the reproducibility knocked on the door accompanied by the time you need to process quite a few images and they are friends of the common ghost called objectivity and you had to write a program.

The challenging point is the varying intensity of the image. In the middle part it gets nearly lost, but the outer part is bright and clear. By thresholding you get an image which only contains the outline, but the inside is relevant as well. So you adjust the threshold, get the inner part and lose the outline to noise... yay. And so on ;) But now it's solved!!

Where the image comes from? Actin is stained with a fluorescent dye and we want to observe the actin activity of sea urchin embryos during cell division :) But that's worth another post, sorry! ;)