I have four images that I plan to make them in equal range and then do classification on them. As I know I should use:
Rater……Data conversion….convert datatype
I used it and then in further processing, I chose ‘peak clipped histogram’. Please look at below image:
I do not know I chose the correct option or not but after converting, three images do not have equal size;
image 1: 0-255
image 3: 23-255
image 4: 1_255
Why is like that? Would you please help me?
the data distribution before the conversion decides how your data is scaled to 8 bit (0-255)
For example, Intensity ranges between 0 and proably 50000 but most values are within a very small low segment. Still, the lowest value will be 0 and the highest 255. All other values are distributed between it. Converting these won’t change the original data distribution but only attach new values to this range.
If you want the values equally distributed (gaussian curve) between 0 and 255 you need to
a) transform the before the conversion OR (eg. log transformation)
b) try a different scaling (log is also available here)
Dear @ABraun Actually I used logarimetric scale and others but I have same problem:
I still have that problem with logarimetric scale even worser.
is it an option to calibrate the data first (Sigma0) and convert it to db? Then it has a gaussian distribution and might look better in 8bit.
There is no option to do it.
Now, my question is this; if values are distributed like below and only a bit difference in values, the classification over them is ok or not?
why shouldn’t it be okay to classify them?