Hello,
I have an image of an animal (mantis shrimp) along with seeding particles (white dots). I want to mask out the animal only and change the background into white. I will use the resultant image as a mask in PIV analysis.
I have tried intensity thresholding, and minimum, median filters but they are of less help because the animal tail has hair with low intensity value and as a result it is also deleted with white dots. I think thresholding based on intensity value doesn't work.
I can simply crop the animal by outlining it but because I have hundreds of images this method is pointless.
I attached the original image along with the desired one done through outlining by hand.
Use mathematical morphology, see top-hat operation (not direct but 1-TH)
Some ideas for you:
- Threshold with Otsu and keep the object with the largest area;
- Threshold + Region growing of the object with larger area.
if you have a few more manually segmented images like the one you show above, one solution is using U-net model for segmentation.
example of implementation:
https://github.com/jocicmarko/ultrasound-nerve-segmentation
However, consider to use other conventional image processing approaches if you are not dealing with very sophisticated images.
Draw a histogram of the pixels. Then locate the bin value that mainly covers the insect defining area. Let this bin in the histogram is BinInc. In the given image, the pixels at the tail and along the border of the insect are close to the BinInc value. Hence, these value containing bins will be the neighbors of BinInc in the histogram. You can form binary image by modifying pixel values of (BinInc to BinInc-R) to 255 and the rest to 0. You can analyse for R value of 1 to 50 to find a suitable R. For better result, apply partitioning the image into blocks and then apply the stated method. Thereafter, apply morphological closing and opening parameters to reduce noises.
By subtraction two sequences image of moving object(if you have video stream) , then extract the object.
Thanks everyone for your help. I tried some of the methods you suggested which are top-hat transform and active image segmentation. Still the methods don't mask entire animal including its tail hair. I tried below steps and found it the best for now. ImageJ was used.
1. Convert images into 8bit grayscale.
2. Apply white top hat with rad=4.
3. Subtract the result of step 2 from the result of step 1.
4. Apply the threshold with relevant intensity. Threshold value is found 18 in my case.
I attached the resultant image.
I think if I can somehow increase the intensity level of the tail hair region before step 4, then that region should also give desired result.
Canny edge detection
Threshold otsu method
Morophlogical operation
Can help u to segment ur object
- Badges
- Science topic
- Similar topics
- Engineering
- Electrical Engineering
- Signal Processing
- Image Processing