edgepreserving

Edgepreserving refers to techniques in image processing and signal processing that reduce noise or small-scale detail while maintaining edges and boundaries between regions. The objective is to suppress unwanted variation without blurring prominent transitions, distinguishing edge-preserving methods from conventional smoothing such as Gaussian blur.

Approaches are often adaptive, using local image structure to control the amount of smoothing. Local methods include anisotropic diffusion (Perona–Malik) and the bilateral filter, which leverages both spatial proximity and intensity similarity to smooth within regions while preserving edges. The guided filter is another efficient local technique.

Non-local and patch-based methods extend this idea by averaging similar patches found throughout the image, as in non-local means and related algorithms. Explicit regularization, such as total variation minimization, frames edge preservation as a constraint in optimization to suppress noise while keeping gradients near edges.

Applications span image denoising, pre-processing for segmentation, HDR imaging, medical imaging, and video processing. Evaluations use objective metrics like PSNR and SSIM, as well as edge-focused measures and qualitative assessment; artifacts such as halos or staircasing may arise with some methods.

Historical context places anisotropic diffusion with Perona and Malik in 1990, bilateral filtering in the 1990s, and subsequent developments including non-local means and guided filtering, alongside data-driven approaches in recent years.