IMAGE DENOISER ENHANCEMENT VIA GENETIC ALGORITHMS

Resumo

Image processing is becoming a more open and extended field, with endless possibilities
and meaningful applications, specially for medical images. There, a point that brings much interest is
the X-ray imaging, which are generated by sending high-frequency electromagnetic radiowaves, and
then retrieving them in a metal plates that they will collide with. But those metal plates might receive
electromagnetic interference or uncontinuous radiowaves, causing image noise, which are inevitable.
That noise can be decreased by image filters. Although those filters have become standard and work
well, their parameters values are static, even though there are many possible parameters values and
each image will have different denoise results with different values. This articles descibres the use of
Genetic Algorithms to find the best parameters with fewer processes. GA are heuristic search
procedures based on natural selection. Only the algorithms that give the best results will generate other
algs, decreasing the number of processes needed. The method uses Genetic Algorithms to find best
parameters according to each image for the “Non-Local Means” denoise filter from the OpenCV
Python library. The X-ray images used in were Optical Coherence Tomography of people with
Pneumonia, published on the online database Mendeley by the University of San Diego California,
and tested on the Python scripts implementations. By the results obtained, it was noticeable the
positive correlation between the filter parameters chosen by the GA for each given image and the
improvement of image denoise from the classical methods, proving that Genetic Algorithms are great
for such applications, decreasing the number of processes needed to find the the best parameters for
any input.