INFUSING HEXAGONAL STRUCTURES WITH MACHINE LEARNING FOR AN IMPROVED IMAGE DE-NOISING SYSTEM

Image de-noising systems require large amounts of computations to successfully de-noise the image with high levels of peak signal-to-noise ratios (PSNR). Hexagonal structure-based representation of images is known to provide a wider range of pixels for effective de-noising. Machine learning is utilized to optimize the quality of service for any kind of computational system. So, in this work, we combine the advantages of both hexagonal structures and machine learning to develop a state-of-the-art image de-noising system. A Fixed window will always have a fixed size of search space, which reduces the efficiency of searching for a non-noisy pixel. If the window size is fixed then there might be redundancy in selecting the non-noisy pixel, thereby increasing the delay of the system. To overcome th above limitation, the proposed system works by adaptively changing the window-size used for de-noising images, and selects the best window size required for de-noising the given block of image data. However, the learning factor incorporates execution delay into the algorithm's design, thereby reducing the overall delay of image de-noising. Results demonstrate that the proposed algorithm improves the PSNR value by more than 20%, while the de-noising time is reduced by more than 15% when compared with the algorithm's non-machine learning counterparts.