Unsupervised Machine Learning Algorithm for MRI Brain Image Processing

Denoising of an image is the first and primary pre-processing step in image processing. In this paper, an algorithm is implemented using machine learning in conjunction with wavelet-based denoising method. Most learning algorithms use activation function that is continuously differentiable. Since standard threshold functions are weakly differentiable, a new type of thresholding function was proposed. Stein's unbiased risk estimate (SURE)-based updating algorithm is used for estimation. The proposed method is compared with conventional filtering andwavelet-based denoising methods, using performance evaluators like PSNR and MSE. Results indicate there is a significant reduction in MSE and increase in PSNR for the proposed method.