A Neural Network-Based Method for Eliminating Stripe Noise

The uneven photoelectric response of the infrared imaging detector system due to the level of craftsmanship results in a fixed deviation in the response of each pixel to the incident light signal. The deviation appears as a stripe noise on the output image, seriously reducing image quality. Infrared images with stripe noise seriously affect application results, therefore the original details of the image need to be effectively preserved by removing all stripe noise artifacts and balancing real-time performance. We proposed a new method for removing stripe noise using the correlation between adjacent column signals to remove independent column stripe noise. Deep neural networks were used to implement this method, using the estimated noise output from the previous network iteration as the input to the next iteration of the network. This approach effectively reduced the search space for function approximation and allowed for effective training on larger datasets. This method can be used to estimate the stripe noise of infrared images and preserve scene details to a large extent. A large number of experimental results showed that the proposed research method has significant effects.