BDNFF - A Novel Bayesian Adaptive Filtering Algorithm for Removing Dynamic Pattern Noise in VHRI Satellite Images

One of the main challenges in remote sensing data is noise, which can negatively impact data quality and analysis results. Pattern noise can have different shapes in the Frequency domain which can be dynamic based on features. FFT thresholding may not suffice because of the difficulty in setting adaptive noise thresholds based on features in the image. There is a need for a methodology that automatically detects and corrects features based on dynamic noise frequency to improve the overall quality of data. This paper aims to explore the various challenges associated with noise removal in Cartosat 2 data, including methods for detecting and reducing noise, and current limitations in feature-dependent pattern noise correction. We propose a novel adaptive noise reduction algorithm for satellite images that can effectively reduce horizontal and vertical components of feature-based dynamic noise. The proposed Bayesian Dynamic Noise Filtering in Frequency domain (BDNFF) method consists of two stages: first, identification of noise was carried out using local statistical characteristics of the data, and second, denoising of the image is done using the Bayesian estimation technique to reduce the identified noise. The effectiveness of the proposed BDNFF method is evaluated using a set of satellite images with various levels and types of dynamic pattern noise. Traditional noise removal methods like using notch filters in the frequency domain introduce information loss due to frequency cutting which the BDNFF algorithm restores using Bayesian estimation. The proposed method has the potential to significantly improve the quality and accuracy of satellite imagery for various remote sensing applications. Analysis and results are presented in the paper in detail. These techniques will help to generate seamless products with a better signal-to-noise ratio (SNR) and modulation transfer function (MTF), which helps in ground segment remote sensing applications.