Textural Region Denoising: Application in Agriculture

Abstract

Geo-science and remote sensing technologies play enormous roles in agriculture nowadays, especially in analysis of data from aerial images such as satellite images and drone images. Most agricultural images contain more textural regions than non-textural regions. Therefore, data management in terms of textural regions is very important. Indeed, additive white Gaussian noise (AWGN) is the fundamental problem in digital image analysis. In wavelet transform, Bayesian estimation is useful in several noise reduction methods. The Bayesian technique requires a prior modeling of noise-free wavelet coefficients. In non-textural regions, the wavelet coefficients might be better modeled by super-Gaussian density such as Laplacian, Pearson type VII, Cauchy, and two-sided gamma distributions. However, the statistical model of textural regions is Gaussian model. Therefore, in agricultural images, we require flexible model between super-Gaussian and Gaussian models. In fact, the generalized Gaussian distribution (GGD) is the suitable model for this problem. Therefore, we present new maximum a posteriori (MAP) estimator for spacial case of GGD in AWGN. Here, we obtained the analytical form solution. Moreover, this research work will also describe limitations of GGD application in Bayesian estimator. The simulation results illustrate that our presented method outperforms the state-of-the-art methods qualitatively and quantitatively.