VLSI Architecture for Decision Tree Based Noise Detector and Gaussian Filter for Noise Removal

This paper focuses on noise removal using a hybrid approach that integrates a decision-tree-based noise detector with a Gaussian filter to reduce Gaussian noise while preserving important image details. Gaussian noise, characterized by random intensity variations, often leads to a loss in visual quality. Conventional filtering applies uniform smoothing, which can cause blurring of edges and textures. To overcome this, a decision tree composed of three modules uniformity analyzer, edge transition detector, and neighborhood correlation evaluator is used to accurately identify noisy pixels. Only the pixels classified as noisy are processed through a Gaussian filter, preserving noise-free regions. Implemented in Verilog and simulated using Vivado, this method leverages parallelism for efficient processing. MATLAB is used for preprocessing and analysis. Results show significant improvements in SNR, PSNR, and SSIM, validating the method’s effectiveness in enhancing image clarity.