A Novel Technique for Removal of High Density White Spot Noise from Digital Neutron Radiographic Images
This paper proposes a novel technique of adaptive switching alternative median (ASAM) filter for high-density white spot noise removal. The ASAM filter is composed of two blocks filtering, namely main and secondary block filtering, respectively. The proposed secondary block filtering is a new technique in high-density impulse noise removal and the main contribution of this research. The ASAM algorithm was tested on the standard 8-bit gray-scale, 512×512 pixel Lena image and a real neutron radiographic image. The results showed significant reduction of white spot noise in both types of images through visual inspection. To measure the performance of noise removal in simulation test we measured the peak signal-to-noise ratio (PSNR), structural similarity (SSIM) index, and denoising time, and in real application tests we measured signal-to-noise ratio (SNR). From the experiments of simulation test, at the highest level noise of 95 % the obtained PSNR and SSIM are 23.584 dB and 0.696 respectively. These are higher than the results of other algorithms that are 16.697 dB and 0.475, respectively, for DBA, 16.696 dB and 0.408 for NAFSM, and 18.860 dB and 0.568 for NASNLM. The denoising times for DBA, NAFSM, NASNLM, and ASAM were obtained as 6.469 s, 5.186 s, 36.735 s, and 5.197 s respectively. From the experiments of real application test we obtained the SNR for DBA, NAFSM, NASNLM, and ASAM as 32.42 dB, 6.01 dB, 18.77 dB, and 32.96 dB, respectively. In general, these results show that ASAM filter is superior to the existing filtering methods. The ASAM filter improved the image restoration quality, especially in removing the high-density white spot noise, and was able to yield good filtering result which exhibits better PSNR, SSIM, denoising time, and qualitative visual inspection.
Copyright (c) 2018 Atom Indonesia
This work is licensed under a Creative Commons Attribution-NonCommercial-ShareAlike 4.0 International License.