Enhanced Switching Median Filter for Denoising in 2D Patterned Textile Images

Abstract

Wallpaper fabric, a part of patterned fabric, is a new decorative material covered with fabric or other natural fibre materials. Wallpaper fabrics are different from other indoor decorative materials and are designed for daily use with easy maintenance instructions. Because of their high demand, quality assurance or fault detection and recognition, in wallpaper fabrics is important. One step that is important during fault detection is the removal of noise. A common type of noise that affects patterned fabric images is the impulse noise. In this paper, a method that modifies a switching median filter is proposed. The algorithm contains two stages, namely, noise detection stage and noise filtering stage. During noise detection a modified adaptive switching median filter with three new thresholds are used to remove the impulse noise. During filtering stage, an adaptive switching median filter is used to determine the size of the sliding window. To detect noisy pixels which are close to its neighbours that are normally missed with the traditional threshold of switching median filter, a modified threshold that considers the ranking order of the pixels in the sliding window is used. Both the enhanced stages improve the noise removal process while maintaining image details. The proposed denoising model was evaluated using three performance metrics, namely, PSNR, FoM and execution time. The experimental results showed that the proposed model is efficient in removing impulse noise with low and high noise density while preserving important image and edge details.