Info iconThis preview shows page 1. Sign up to view the full content.

View Full Document Right Arrow Icon
This is the end of the preview. Sign up to access the rest of the document.

Unformatted text preview: imizing the Poisson noise. 19.05.2009 Dr. Pierre Elbischger - MIP1/ISAP'SS09 43 Increasing the SNR by low pass filtering Major assumption: Neighboring pixels have the same gray value, or a least differ only slightly. A N×N uniform filter kernel improves the SNR by a factor of N (the square root of N²) Due to the low pass filtering the resolution of the image is decreased. The strategy can be continued until the filter kernel is equal to the size of the object being detected. This means the ability to detect an object is proportional to the square-root of its area. If an object's diameter is doubled, it can be detected in twice as much noise. Original image 19.05.2009 3x3 low pass filtered Dr. Pierre Elbischger - MIP1/ISAP'SS09 11x11 low pass filtered 44 Increase the SNR by ensemble averaging Major assumption: pixel readings at different times represent the same structure in the viewed scene. Fluorescence light microscopy images original size close-up 19.05.2009 one image Dr. Pierre Elbischger - MIP1/ISAP'SS09 ensemble average 45 E12 Increase the SNR by ensemble averaging – Integration over time SEM images of a scratched metal surface The distribution becomes narrower → decrease in the variance of the noise 1 second scan 19.05.2009 2 second scan Dr. Pierre Elbischger - MIP1/ISAP'SS09 46 Low-pass filters vs. order filters Low-pass filtering works fine in the case of Gaussian-like noise but is not effective in removing salt & pepper noise. Single pixels that show a very different gray value influence the mean of their neighborhood. Original Salt and pepper noise Mean filtering Solution: Ranking of the pixels in a certain neighborhood according to their gray value. Then, for example, the median, minimum or maximum value can be used as the gray value for the center pixel. Benefits • Produces no new gray values • Retains edges in the image 19.05.2009 Dr. Pierre Elbischger - MIP1/ISAP'SS09 47 E18 Order filters - Median filter • Retains edges in the image • Slow because of its non-linearity salt & paper noise (impluse noise) original 3...
View Full Document

This note was uploaded on 07/09/2009 for the course MEDIT 1 taught by Professor Pierreelschbinger during the Spring '09 term at Carinthia University of Applied Sciences.

Ask a homework question - tutors are online