I need help for the following problem. Suppose that I have two images of a patient‘s body one today and an older one a couple of years ago. The problem is: find the aerias of the skin where moles have changed color or shape. I am a newbie on computer vision so I am asking for quidance from where to start. Are there any algorithms known to solve this problem or ca you suggest some papers or algorithms to start from??
Most standard computer vision approaches would include matching the images first and then comparing their characteristics. Basic and common methods would be feature matching, template matching, available in most computer vision libraries such as OpenCV (googling that will give you lots of details). More advanced techniques, especially in medical imaging, would feature neural networks, specifically convolutional neural networks. They could be used for segmentation of lesions from your photos and then you could easily compare shapes (assuming that two images are properly matched/registered).
Then again, question is what is your goal. It seems like you're going for melanoma prediction, and state of the art shows that this sort of "feature engineering" you're suggesting is not efficient. We are able to create human-level classifiers without directly&quantitatively analyzing colors or shapes. Now famous 2017 paper by Stanford's AI Lab showed human-level skin cancer detection with deep learning approach:
https://www.nature.com/articles/nature21056
https://cs.stanford.edu/people/esteva/nature/
I think this is a problem could be solved by using Gabor filter + some techniques like PCA or K-MEANS.
I recommend the following link to you:
https://www.mathworks.com/help/images/texture-segmentation-using-gabor-filters.html
Identifying the skin pixels from the images based on skin tone color is done using skin tone algorithm. Color space transformation is used to detect and trace the presence of skin pixels in an image. The adjustment of skin tone values is done according to the requirement taking care that no environmental or background effects identifying the pixels. Usually skin tone algorithm is done on videos for the fact that embedded data will be spread on the entire image. While reconstructing the image reference point required will be lost due to scaling, rotation or cropping of any other video processing technique. This disadvantage is overcome when skin tone algorithm is employed which makes use of transformed color space.
Refer the following Link:
These links may be useful for you.
http://shodhganga.inflibnet.ac.in/bitstream/10603/92404/11/14.chapter%205.pdf
http://www.alliedacademies.org/articles/study-and-development-of-algorithm-of-different-skin-diseases-analysis-using-image-processing-method.html
https://pdfs.semanticscholar.org/fee0/08af8cb3f9636c82cf8343c078fd6715bd59.pdf
- Badges
- Science topic
- Similar topics
- Papers