Below i have attached an original image with segmented output. I need to generate an exact mask of the original image which should have proper boundary with interior full with white color intensity pixel(just like an template).
If there is any existing algorithm, deep learning model or paper available, kindly suggest.
You can do this easily with morphological operations. First close the image to close cracks on the image. Then apply tophat transform to normalize the unbalanced light effects. Then open and close for a compact mask. Here is a sample code for matlab.
imb1=imclose(img,strel('disk',5));
imb1= imtophat(imb1,strel('disk',50));
imb2=imopen(imb1>40,strel('disk',5));
imb3=imclose(imb2,strel('disk',20));
Further to Abdullah H. Ozcan 's answer, you could look at using a "Sobel operator" and then do some post-processing to fill in the cracks.
You may find the following article interesting:
Article A Modified Segmentation Approach for Real World Images Based...
In terms of image processing, morphological operations can be an option as suggested above. But before applying them, you may want to look at Otsu's method (matlab code/function available) for thresholding to convert the grayscale image to a binary one, then the outlier white/black portions can be removed using morphological opening/closing operation as suitable. Another thing you can try before applying morphological operations is detecting the boundary, that will help a lot i think--
https://www.mathworks.com/discovery/edge-detection.html
In terms of deep learning, there are lots of image segmentation network architectures available which you may use (though your case seems much simpler for that). You can simply to train a CNN with your images with to train, I'm sure it'll produce good results, for ref--
https://blog.goodaudience.com/using-convolutional-neural-networks-for-image-segmentation-a-quick-intro-75bd68779225
Hope this helps,
Sayeed
- Badges
- Science topic
- Similar topics
- Engineering
- Electrical Engineering
- Signal Processing
- Image Processing