This paper presents a two-stage impulse noise removal filter from medical images. Quaternion is used to represent differences of two pixels. The pixels are sorted and assigned a rank based on the aggregated sum of pixel differences with other pixels inside the filtering window. The central pixel is considered as corrupted by an impulse if its rank is bigger than a predefined rank and the minimum difference between it and other pixels in the four edge direction inside the window is larger than a predefined threshold. The noisy pixel is replaced by output of vector median filter implemented using quaternion. For color images, both intensity and chromaticity components are used. Quaternion processes color images as single unit rather than as separated color channels. This preserves the correlation and three dimensional vector natures of the color channels. For grayscale medical images, the same algorithm is implemented by using the intensity difference between two pixels. Experimental results show improved performance of the proposed filter in suppressing the impulse noise while retaining the original image details comparing against other well-known filters.
This study presents a novel two-stage filtering algorithm for removing impulse noise in color images. Quaternion theory is used to represent the intensity and chromaticity differences of two color pixels. Use of quaternion treats color pixels as vectors and processes color images as single unit rather than as separated color components. This preserves the existing correlation and three dimensional vector natures of the color channels. In the first stage of noise detection, the color pixels are sorted and assigned a rank based on the aggregated sum of color pixel differences with other pixels inside the filtering window. The central pixel is considered as probably corrupted by an impulse if its rank is bigger than a predefined rank. In the second stage, the probably corrupted candidate is again checked for an edge or an impulse by using four Laplacian convolution kernels. If the minimum difference of these four convolution is larger than a predefined threshold, then the central pixel is regarded as an impulse. The noisy pixel is replaced by output of weighted vector median filter implemented using the quaternion distance. More weight is assigned to those pixels belonging to the direction of minimum difference. Experimental results indicate the improved performance of the proposed filter in suppressing the impulse noise while retaining the original image details comparing against other well-known filters.
scite is a Brooklyn-based organization that helps researchers better discover and understand research articles through Smart Citations–citations that display the context of the citation and describe whether the article provides supporting or contrasting evidence. scite is used by students and researchers from around the world and is funded in part by the National Science Foundation and the National Institute on Drug Abuse of the National Institutes of Health.