It will depend on your database. You can start with traditional computing, extracting resources, such as filters, lbp, surfing, etc... and if the result is not what you expected, you can try methods that require more processing, such as artificial neural networks and finally deep learning, if you have a robust database and gpu.

An example image would be helpful. If you want to determine the length of something string-like, following equivalent pixels comes into mind. Maybe gaps have to be bridged, then splines may be useful.

Regards,

Joachim

Wanderson Gonçalves , I have tried to use filters, but they did not work, since the collagen bundles are closely dense and tend to change their orientation depending on the disease type

Joachim Pimiskern The collagen bundles, sometimes are densely packed and their orientation changes , but sir in the attachment you can find an example , (I want to detect and extract the length of the black lines )

The first step should be gluing areas together. Can be done with a Gaussian filter. Areas are pixels that are transitively connected by similar brightness. I'd use the flood fill algorithm to collect them. floodfill(...) is available in OpenCV, you may also have a look at what OpenCV says about blobs.

The next step is measuring the lengths. Ideas might be: go through all pairs of pixels of an area and determine the maximum Euklidean distance between them, or apply a shortest path search, then take the longest shortest path as measure.

Attached is a preprocessed sample (manually, with Paint.NET). It indicates how melting pixels together might help. Finding the best Gaussian kernel, the number of brightness levels, are optimization tasks on their own.

Regards,

Joachim

Joachim Pimiskern Thank you very much sir for your help. Sir can i apply those methods for a tiff image (it includes 850 pic) without performing a Z projection ?. Also, what is the best method to obtain the angle of those bundles ?