Why is the bit depth 8 bit for Haralick textures?
The eCognition documentation states that 8 bit data are used for Haralick texture calculations and that the grey-level co-occurence matrix is 256 X 256. Most software I've used uses 16 or 32 grey levels (4 or 5 bit). I would only be concerned about processing time, except that I've read a few sources saying that too many grey levels can cause poor estimates of some of the underlying statistics (see p. 17 here https://prism.ucalgary.ca/handle/1880/51900 and this article https://www.tandfonline.com/doi/abs/10.5589/m02-004). Does anyone know why the developers chose 8 bit? Are there other supporting studies I don't know about?
Was this article helpful?
1 comment
Hello Andy,
Good question! (ˇ෴ˇ)
256 grey levels (8 bit) were usually at the time we implemented the Haralick features. I am not sure if the reduction to 4 or 5 bit (16 or 32 grey levels) will increase the precision, but I am not an expert in this area. (⊙_☉)
Please insert an request to our software wish list to add an parameter to the Haralick features to define the grey levels (4, 5, 6, 8 bit) by the user: https://support.ecognition.com/hc/en-us/community/topics/360000772240-Software-Wish-List?sort_by=recent_activity
Cheers (ツ)
Christian