ST bilateral-deep filter: Shearlet transform based bilateral filter and deep learning approach for noise reduction in CT images

This paper develops a ST-BF+Taylor-ACVO for effective denoising. The results from ST and deep learning are combined using the fusion-based quality measure to compute the restored image. Input noisy image is passed to CNN, where noise pixels are identified and the pixel restoration is done by proposed Taylor-ACVO. To produce a denoised image, the pixel enhancement is done using the vectorial total variation norm. On the other hand, input noisy image is applied to ST and the resulted image is fed to bilateral filter to generate noise free image, which is applied to Inverse Shearlet transform to reconstruct back the original image. Finally, the image obtained from VTV norm and IST is fused by the quality metrics to compute restored image. The proposed method obtained higher efficiency in terms of PSNR, SDME, and SSIM with values of 27.85 dB, 40.27 dB, and 0.87 using Gaussian noise.