Reduction of random-valued impulse noise by using multi-structured textons

In this paper, an iterative two-stage image denoising technique based on multi-structured textons for noise identification while spatially linked directional similarity for noise restoration (MTNI-SDSNR) is presented for the denoising of random-valued impulse noise (RVIN). Multiple textons oriented at various directions of a sliding window are proposed for the identification of noisy pixels, via an adaptable threshold range computed from their local statistics. Whereas the spatially similar noise-free pixels in the 4 spatially linked directional neighboring pixels obtained by computing the local similarity among them, are used for noise restoration. As the textons are elementary for texture perception, so they are proposed to be oriented based on reflectional symmetry to ensure the effective preservation of salient edges. The proposed MTNI-SDSNR is compared with state-of-the-art denoising methods using standard benchmark grayscale and biomedical images taken from MedPix dataset, by corrupting them with various RVIN intensities. The supremacy of proposed method with similar benchmark RVIN denoising methods can be depicted in quantitative results by showing an increment of 2% (on average) in the values of Peak-signal-to-noise-ratio (PSNR), structural similarity index measurement (SSIM). The visual results represent the edge-preserving capability for both lower and higher intensities of random noise.