Salt-Pepper Noise Removal Approach Based on ANN and Directional Mean Filtering Algorithm

To preserve the image details more precisely and fully after image filtering, a directional mean filtering algorithm based on the edge direction of noise points is proposed. Firstly, pepper and salt noise recognizer model based on single hidden layer neural network is established. Afterwards, pepper and salt noise recognizer is trained to detect noise points, respectively. Then, according to the neighborhood pixel value of the noise point, the direction with the smallest pixel difference is regarded as the edge direction. Finally, the pixel value of the noise point is replaced by the average value of the pixel in the edge direction. The experimental results indicates that the filtering effect of the algorithm is superior to other algorithms on low and medium noise concentrations, with a PSNR value up to 82.0dB. While for the high concentration of noise, the PSNR value and consumed computational time are less than 3.2dB and 3.91s, respectively. According to the experimental results, the proposed algorithm provides a better capability of filtering and displaying detailed information of the image, and significant guidance on promoting the construction of the energy Internet and digital grid.