Digital Signal Processing 18 (2008) 406–421 www.elsevier.com/locate/dsp An impulsive noise color image filter using learning-based color morphological operations Hao Zhou ∗ , K.Z. Mao School of Electrical and Electronic Engineering, Nanyang Technological University, 639798 Singapore Available online 13 May 2007 Abstract Morphological filter is a kind of morphological operation-based nonlinear filter. It is effective in impulsive noise filtering and has been extensively studied in the past two decades. In the presented study, a new multichannel filtering approach established on learning-based color morphological operations for impulsive noise removal in color image is presented. By using the color pixel ordering scheme learned from the pre-estimation of impulsive noise, contaminated pixels are ordered as maximum ones in erosion operation or minimum ones in dilation operation, respectively. This characteristic ensures that only uncontaminated color pixels are distributed in morphological operations, hence noisy pixels are suppressed. Reconstruction is followed to alleviate the blurring and bias effects of morphological operations and to preserve image features. The presented filtering approach greatly enhances the performance of morphological operation-based filters, especially in the color image highly corrupted by impulsive noise. Experiments and comparisons with classical filters, such as basic vector median filter (VMF), basic vector directional filter (BVDF), NOPNCP filter, etc., as well as some newly developed filters, are performed to demonstrate the effectiveness of the proposed color image filtering algorithm.  2007 Elsevier Inc. All rights reserved. Keywords: Impulsive noise filtering; Color image processing; Mathematical morphology 1. Introduction Impulsive noise is often encountered during image transmission process [1]. The presence of impulsive noise in an image will introduce errors in the information acquisitions and produce undesirable effects in the subsequent image processing steps [2]. In general, optimum reception schemes designed for Gaussian noise environments perform very poorly when impulsive noise is present [3]. As a result, filtering the noisy image to suppress the noise and meanwhile preserve the image features, such as edges, textures and other details, is a necessary and significant pre-processing step before any image processing applications. Nowadays, color image is the dominant image format, and color image denoising has attracted increasing atten- tions [2]. The most convenient way to filtering color image is to directly extend conventional grayscale image filtering approaches [4–8] to each dimension of the color image. However, performing noise filtering in each channel sepa- rately might introduce new color artifacts in resultant images. Therefore, the major efforts in color image filtering are * Corresponding author. E-mail address: zhou0020@ntu.edu.sg (H. Zhou). 1051-2004/$ – see front matter  2007 Elsevier Inc. All rights reserved. doi:10.1016/j.dsp.2007.04.013