I think it would be better with 256 intesity levels...Ostu's method is used to find the global threshold in the image and he does it by calculating the probability that each level has of repeating itself.....so with more intensity levels you will have more information about the image....i think 4 levels is not enough with 4 levels your image will be practicaly binarized and the loss of information will be huge.....anyway that's my opinion...greatings

Ostu's method tries to find the threshold that minimizes the intra-class variance of the two populations (pixels lower than the threshold, and the ones higher or equal to it).  As such, if you have 8bit image, you should use all 256 values. Otherwise, you will sacrifice precision by having threshold values restricted to something like (64, 128, 192) instead of the "best" value between 0-255.

Hi  Luis Alberto Rodríguez Rodríguez and Bruno Martin.

Thanks.

I got better result when my images have 4 intensity levels!!!