I am working on binary segmentation of ultrasound images, i.e. from a noisy (mostly speckle noise, I would say) ultrasound image I want to obtain a black and white image.
For finding the global pixel intensity threshold I'm using Otsu's method, which works reasonably well.
Now my question is, what image (pre-) processing steps would one perform on an ultrasound image prior to finding the intensity threshold via, e.g. Otsu's method.
Is Gaussian blurring, median filtering etc. a good idea or does it destroy the pixel intensity histogram? What is the common work flow or image processing pipeline prior to image segmentation?
Thanks a lot for your insights (I'm new to image processing)
The ultrasound image is a transcranial Doppler spectrogram to assess blood flow velocity of cerebral arteries.
I wrote an algorithm for tracing the maximum velocity (envelope) which works quite well. Now, I'm trying to improve the algorithm in terms of robustness and efficiency.
My basic workflow is to take the image ultrasound image, apply a median filter to reduce speckle noise, then I find a threshold via Otsu method and on the black and white thresholded image I find the max. contour, which is the "envelope" traced in red.
By an elongated kernel for the median filter you mean choosing for the median the neighbourhood [m n] size such that m>n, right?
- Badges
- Science topic
- Similar topics
- Gels