Ananias Pereira Neto ^1* Fabrício J. B. Barros ²

• ¹ Signal Processing Departament, Federal Institute of Education, Science and Technology of Pará, Belém, Brazil
• ² Graduate Program in Electrical Engineering, Federal University of Pará - UFPA, Belém, Brazil

Wavelet thresholding techniques, which adjust wavelet coefficients, are essential to mitigate or eradicate unwanted distortions in data communication systems. This is particularly important in computer applications and digital storage, where various interferences, primarily noise, can alter the information at different stages of the process. Noise reduction in images has become an important step in improving the visual quality of signals. The effectiveness of wavelet transformbased thresholding functions is vital for enhancing image quality, as they typically result in fewer edge and texture artifacts, providing more uniform noise reduction. Several thresholding functions have been proposed to improve noise reduction in images. However, some existing methods often present issues such as missing edges and textures, poor smoothness, discontinuous functions, and the need for parameters determined through trial and error. To address these challenges, this study proposes a new method for noise reduction in brain magnetic resonance imaging (MRI).The proposed approach uses an adaptive wavelet thresholding technique that selectively reduces or eliminates noise wavelet coefficients deemed irrelevant to the processed image. The threshold is adjusted based on a linear prediction factor, which exploits the correlation between the original and noise images. The linear prediction factor utilizes temporal information from the images, along with features from both the noise and original versions, to compute a weighted threshold. This threshold is subsequently applied in the wavelet thresholding function to refine the wavelet coefficients, leading to a more efficient noise reduction. The proposed method was evaluated against state-of-the-art noise reduction techniques. Experimental results show that it delivers significant improvements in key metrics such as mean squared error (MSE), peak signal-to-noise ratio (PSNR), and structural similarity index (SSIM).