Image Restoration Using Noisy ICA, PCA Compression and Code Shrinkage Technique

The research reported in the paper aims the development of some methodologies for noise removal in image restoration. In real life, there is always some kind of noise present in the observed data. Therefore, it has been proposed that the ICA model used in image restoration should include noise term as well. Different methods for estimating the ICA model when noise is present have been developed. In noisy ICA, we have to deal with the problem of estimation of the noise free realization of the independent components. The noisy ICA model can be use to develop a denoising method, namely the sparse code shrinkage [10]. The final part of the paper presents a LMS optimal PCA compression/decompression scheme, where the noise is annihilated in the feature space. In order to derive conclusions concerning the correlations between the dimensionality reduction and the resulted quality of the restored images as well as the effect of using both LMS optimal compression/decompression technique and PCA based noise removal method several tests were performed on the same set of data. The tests proved that the proposed restoration technique yields high quality restored images in both cases, when the CSPCA algorithm was applied directly on the initial image and when it was applied in the reduced feature space respectively.