Abstract:
Hidden Markov model in wavelet field has lately become valuable technique for image processing. Hidden Markov tree model holds ability to capture the significant attributes of wavelet coefficients of data. In image processing, denoising of images that are corrupted with Gaussian noise is a well-known problem. This research work makes an attempt to solve this problem by proposing a wavelet based technique. We propose an Expectation-Maximization algorithm used on 2D Discrete Wavelet Transform of experimental data; DWT of data is modeled by HMM which captures the non-Gaussianity of wavelet coefficients. Proposed technique outperforms few prevailing methods with regards to Peak Signal to Noise Ratio (PSNR). Various thresholds are used on convergence error in EM algorithm to analyze the effect and optimize the selection of threshold on basis of results.
