GENETIC FUZZY FILTER BASED ON MAD AND ROAD TO REMOVE MIXED IMPULSE NOISE

In this thesis, a genetic fuzzy image filtering based on rank-ordered absolute
differences (ROAD) and median of the absolute deviations from the median (MAD) is
proposed. The proposed method consists of three components, including fuzzy noise
detection system, fuzzy switching scheme filtering, and fuzzy parameters
optimization using genetic algorithms (GA) to perform efficient and effective noise
removal. Our idea is to utilize MAD and ROAD as measures of noise probability of a
pixel. Fuzzy inference system is used to justify the degree of which a pixel can be
categorized as noisy. Based on the fuzzy inference result, the fuzzy switching scheme
that adopts median filter as the main estimator is applied to the filtering. The GA
training aims to find the best parameters for the fuzzy sets in the fuzzy noise
detection.
From the experimental results, the proposed method has successfully removed
mixed impulse noise in low to medium probabilities, while keeping the uncorrupted
pixels less affected by the median filtering. It also surpasses the other methods, either
classical or soft computing-based approaches to impulse noise removal, in MAE and
PSNR evaluations. It can also remove salt-and-pepper and uniform impulse noise
well.