Multistage Filtering Algorithm for Salt and Pepper Noise Removal from Highly Corrupted Microscopic Blood Images

Abstract

Image quality mainly affects the performance of any
medical image processing system. Salt & pepper noise is one
type of noise that significantly degrades the image quality. Such
noise can be frequently encountered in digital microscopic
images due to technical reasons. Moreover, high ratios of Salt &
pepper noise make the image excessively corrupted or
unreadable. Standard and modified median filters can usually
handle low/medium Salt & Pepper noise densities, mostly at the
expense of edge/details preservation. However, they totally fail
for highly corrupted images where noise density reaches 90%.
In this paper, we present a new multistage filtering algorithm
for Salt & pepper noise reduction from highly corrupted images
while preserving image details and edges as better as possible.
The proposed algorithm includes two filtering stages through
which image is firstly de-noised via utilizing adaptive median
filter then decision based median filter. Our multistage filter
has been successfully applied on noisy microscopic blood images
obtained from Malaria-infected blood smears. Results reveal
that the presented filtering algorithm outperforms standard and
modified median algorithms in terms of PSNR, MSE and IEF
values, specifically for images with more than 80% of salt &
pepper noise. This indicates that using our multistage filtering
algorithm against high Salt & Pepper noise densities, does not
only remove the noise effectively but also achieves a better edge
and details preservation, hence a better image enhancement.