Entropy Based Parameter Estimation of 2D Gaussian Filter for Image Speckle Noise Removal

Document Type : Research Article

Authors

1 Department of Biomedical Engineering, Faculty of Electrical Engineering, K. N. Toosi University of Technology

2 Department of Electrical Engineering, Imam Khomeini International University, Qazvin, Iran

Abstract

In this paper, a speckle noise suppression algorithm based on the 2D Gaussian filter is addressed, which employs entropy to estimate the filter's variance effectively. Speckle noise is an inherent characteristic of coherent imaging systems which degrades the quality of resulting images. Gaussian filter is a traditional approach for speckle denoising; however, estimating its optimum variance is still a challenge. Many algorithms have been developed to estimate the optimum variance but they suffer from the type of noise or a predetermined variance. Our proposed method demonstrates an improved 2D Gaussian filter since it estimates the optimum variance of the filter in the context of differential entropy between the noisy and filtered images under different p-norms. This optimum variance is directly estimated from the speckle noise level of image and it differs for different types of noise and images. The optimization problem is numerically solved, and also, the value of the norm order is appropriately determined. The blind estimation of norm order is also accomplished based on the level of noise variance. Finally, the proposed method's performance is appraised, utilizing both standard and real ultrasound (US) images. The quality of filtered images is assessed through the qualitative and quantitative simulations in terms of peak signal to noise ratio (PSNR), correlation coefficient (CoC), structural similarity (SSIM), and equivalent number of looks (ENL). The experimental results reveal the proposed method's proficiency in contradiction to state-of-the-art despeckling methods through the capability of strong speckle noise removal and preserving the edges and local features.

Keywords

Main Subjects