Gaussian modulated hyperbolic tangent high pass filter for edge detection in noisy images

TL;DR

This paper introduces a novel non-separable 2D Gaussian modulated hyperbolic tangent high pass filter that enhances edge detection in noisy images by improving noise robustness and directional selectivity, outperforming traditional filters.

Contribution

The paper presents a new non-separable 2D high pass filter with Gaussian modulated hyperbolic tangent response, offering improved noise resilience and directional selectivity for edge detection in noisy images.

Findings

3dB PSNR improvement over Laplacian of Gaussian filter

Enhanced noise robustness in high pass filtering

Better directional selectivity in edge detection

Abstract

In this paper, a non-separable (NS), robust to noise, Two Dimensional (2D) isotropic Gaussian Modulated Hyperbolic Tangent (GMHPT) High Pass (HP) filter is designed to filter the high-frequency components present in a noisy image. The major drawbacks of conventional HP filters used for image analysis are noise influences (more sensitivity to noise) which lead to spurious responses and limited directional selectivity due to their separable property. The designed filter employs GMHBT frequency response to ensure improved noise performance in the least square sense. The 2D NS filter has better directional selectivity and GMHBT profile offers less noise sensitivity along with regularization by least square error design. Improvement of high pass filtering in noisy images, by means of restoring the high-frequency components of the image is measured with Peak Signal to Noise Ratio (PSNR) and compared with two dimensional, non-separable isotropic Laplacian of Gaussian (LoG) filter. The designed filter gives 3dB improvement in the PSNR.