An Efficient Curvelet Bayesian Network Based Approach for Image Denoising

The development in the processing capabilities of electronic devices directed the research of efficient image denoising technique towards the more complex methods which utilizes the complex transforms, functional analysis and statistics. Even though with the sophistication of the recently developed techniques, most algorithms fails to achieve desirable level of performance. Most algorithm fails because the practical model does not matches the algorithm assumptions taken at the time of development. This paper presents an efficient approach for the image denoising based on curvelet transform and the Bayesian Network. The proposed technique utilizes the statistical dependencies in the curvelet domain to train the Bayesian Network which is then used for predicting the noise probability. The curvelet transform provides better approximation especially in directional discontinuities which makes it preferable for processing the pixels around the edges. The experimental results show that the proposed technique outperforms wavelet based methods visually and mathematically (in terms of peak signal-to-noise ratio (PSNR)).