SMART REAL TIME ADAPTIVE GAUSSIAN FILTER SUPERVISED NEURAL NETWORK FOR EFFICIENT GRAY SCALE AND RGB IMAGE DE-NOISING

摘要

Digital imaging may be corrupted by random variations in intensities due to external interferences or noise. Some common models of noise are similar to the real one such as: Salt and pepper, impulse, and Gaussian noise. These distortions may alter the perception or the interpretation of the processed data. They can also cause problems for post processing tasks such as patterns recognition, object detection and medical decisions. In this paper, a new and efficient method for grayscale and RGB image de-noising is presented. Neural networks are used to transform static Gaussian low-pass filter to dynamic smart filter that targets and eliminates different kinds and densities of noise in the image. Simulation results prove that the proposed method is able to filter efficiently corrupted data and reduce noise as well as preserve edges and forms. Applied on grayscale and color image, it overcomes the constraints of the static nature of the Gaussian core. The filtering strength varies with respect to the image characteristics. A comparison with the static filter is conducted to highlight the improvement allowed by this method.