Two-dimensional multi-pixel anisotropic Gaussian filter for edge-line segment (ELS) detection

• A multi-pixel anisotropic Gaussian filter is proposed for edge detection.
• The aim is to detect multiple edge pixels or edge-line segment within a window.
• The filter fully takes advantage of structural information of a boundary.
• The proposed filter can achieve better performance than several existing edge-detection methods.

In order to further improve the performance of the existing anisotropic Gaussian filters and more fully take advantage of structural information of a boundary, we heuristically develop a new multi-pixel anisotropic Gaussian filter to detect edges or edge-line segments directly from low signal-to-noise ratio images. To significantly increase computational efficiency, the classical isotropic Gaussian filters are first used for quickly estimating an approximate direction along an edge; then our filter is applied to more accurately search edge-line segment direction by a few directional filter masks only near such approximate direction. By comparing the proposed filter with the isotropic Gaussian filters, we analyze two improvement factors associated with the localization and SNR of the proposed filter. Experimental results show that the proposed detector can achieve better performance than several existing edge-detection methods in the sense of noise reduction, good localization, and high edge continuity.

Figure optionsDownload high-quality image (60 K)Download as PowerPoint slide