1
$\begingroup$

I wanted to consult with some experts whether or not this pre-processing step is excessive for what I'm trying to achieve. I am in a similar scenario as the author of the article, where I have many many colonies that I'm trying to segment and distinguish from the background. So I found this article:

A System for Automatic Cell Segmentation of Bacterial Microscopy Images

Preprocessing ->

  1. Power-Law Transformation,
  2. Homomorphic Processing,
  3. Log of Image,
  4. Subtracting Original image by Gaussian Kenrel of Image,
  5. Vincent Dome Transform (never heard of this, so I just used a different de-noising algorithm)
  6. Morphological Operator,
  7. 3x3 Median Filter

Then we continue with the main part of the recipe:

II. Background Segmentation

III. Cell Segmentation

IV. Convex Hull Segment Splitting

V. Post Processing

I have no problem going through with this I've already Implemented step 2-6 (not sure how to apply 1). My only concern that it's time consuming to iterate one test for every time I change a value of a controllable variable.

Update

Finally got to ask my professor (an expert of Computer Vision and Machine Learning) and she goes "no this is not excessive at all"

| improve this question | |
$\endgroup$
1
$\begingroup$

I cannot access the raw images, so I did not test the algorithm. Note the aim of pre-processing is to increase the cell contrast to the background. You can remove one step from 2-6 each time to observe the background segmentation effect. By this means you may determine what steps play critical roles on the accurate cell boundary extraction and what others are not necessary. Mostly morphological operations plus sobel filters should be sufficient.

| improve this answer | |
$\endgroup$
  • $\begingroup$ I like that idea, I also decided to cut smaller square of the big picture to start with. The context is cell segmentation, with a ton of colonies, so it's like cutting the sample size to 300 to 20 =] $\endgroup$ – Iancovici Jan 23 '14 at 13:15
0
$\begingroup$

It is not necessary to use all the mentioned pre-processing steps in your code. Go for the goal of the pre-processing if you can achieve your goal with out some of the steps then omit them.

Optimizing the code can reduce the time consumed in iterations.

| improve this answer | |
$\endgroup$
0
$\begingroup$

Chances are great that all this preprocessing is overkill. And also possibly unsufficient in your case. It all depends on the quality of your images.

First check how your segmentation behaves without any preprocessing.

| improve this answer | |
$\endgroup$

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.