An improved local Laplacian filter based on the relative total variation

Affiliation of Author(s):电子信息工程学院

Journal:DIGITAL SIGNAL PROCESSING

Key Words:Edge-preserving image smoothing Local Laplacian filters Relative total variation Image pyramids

Abstract:Local Laplacian filter is an edge-preserving image filter which can smooth image details and preserve image edges very efficiently. In the filtering process of the local Laplacian filter, a simple criterion is used to distinguish large-scale edges from small-scale details. A global threshold is used in the criterion. The intensity distance of a pixel in the input image is defined as the absolute value of the difference of the intensity of that pixel between a reference value. Those pixels with intensity distances smaller than the threshold are considered as small-scale details while those pixels with intensity distances larger than the threshold are considered as large-scale edges. However, this criterion can make wrong decisions when dealing with strong textures or weak edges. Pixels belonging to the strong texture region can have intensity distances larger than the threshold while pixels belonging to weak edges can have intensity distances smaller than the threshold. In such situations, textures which should be smoothed are considered as large-scale edges and they are preserved. Weak edges which should be preserved are considered as small-scale details and they are smoothed. In this paper, this criterion is improved by introducing the concept of the relative total variation. The global threshold is replaced by a local threshold. The local threshold is dependent on the relative total variation and it can make right decisions when dealing with strong textures or weak edges. Experimental results show that, compared to the original local Laplacian filter and several other state-of-the-art edge-preserving image filters, the proposed method can get significantly better image smoothing results especially when there are strong textures in the input images. (C) 2018 Elsevier Inc. All rights reserved.

ISSN No.:1051-2004

Translation or Not:no

Date of Publication:2018-07-01

Co-author:宋昱

Correspondence Author:wyq

Date of Publication:2018-07-01