Temporal G-Neighbor Filtering for Analog Domain Noise Reduction in Astronomical Videos

Abstract

Astronomical images obtained from existing cameras are subjected to various types of noise artifacts. Impulse noise is one of them and it is visible as dark and bright spots on the image. Common practice to remove impulse noise is to perform averaging of several frames. This will increase signal-to-noise ratio of the image; however, impulse noise might still be present. The noisy image will hinder the performance of further processing operations such as edge detection. In this thesis, variable pixel G-neighbor temporal filtering circuit is proposed to improve the quality of astronomical images with the impulse noise. As compared to conventional averaging of frames in time domain, where all pixels are summed, proposed circuit select pixels in each frame that are closest to reference pixel. The circuit operates in analog domain, and it was designed and tested using 180 nm CMOS technology. Simulations demonstrate improvement in peak-signal-to-noise (PSNR) ratio, mean squared error (MSE) and structural similarity index measure (SSIM) as compared to conventional averaging of frames and existing edge-aware denoising algorithms.