RGB image de-noising using new low-pas filter with variable Gaussian core real time optimized by neural networks

Filtering consists in applying a non linear transformation on the image intensities by convolution to modify its characteristics. Gaussian filter is widely used in literature as a low pass filter for signal de-noising. It has some advantages and many inconvenient. It presents a static shape that convolves uniformly the entire image zones. Its smoothing efficiency depends on the value of its standard deviation. More its smoothing efficiency is increased more the image is blurred and the details and borders are removed. All these inconvenient are related to the static nature of Gaussian core of the filter. In this paper we propose a new approach for RGB images filtering, based on a smart dynamic filter with variable Gaussian core based neural network. The parameters that intervene in the filtering process are real time computed and supervised by a neural network. The filter is continuously varied to detect and clean noisy zones and avoid clean zones in the image. The experimental results demonstrate the efficiency of the proposed technique. The image is well filtered and the details and borders are more conserved.