An Approach to Image Denoising Using Autoencoders and Spatial Filters for Gaussian Noise

At a prime age in data science where information extracted from a large variety of data sources drives a lot of research, the need for having clean data is important. Accumulation and extraction of pure and specific data from such sources is an important stage in this area. It is important to consider noise from the data sources in order to pre-process the data. The work focuses specifically on handling Gaussian noise in acquired images that are used in specific domains like image classification. The work proposes a denoising pipeline involving spatial filters and autoencoder networks to remove Gaussian noise in image data to achieve high performance in computer vision tasks. Image data that is abundantly available can be put to good use by the proposed denoising pipeline which enhances machine interpretation of such data. The work provides an empirical analysis on the results obtained and discusses important trends observed.