Single Image Rain Streaks Removal and De-noising Using Self Learning Technique

Image decomposition is an efficient research area for wide variety of applications in the field of image de-noising, image compression, image restoration etc. The main drawback of the prior art algorithms is that, it require training image in advance in order to compute the relationship between input and output dictionaries. In this paper, image decomposition is done with the help of self-learning. This technique recognize image components based on similar semantic features can be used in the applications like rain streaks removal, gaussian de-noising etc. This approach decompose the image into high frequency part(HF) and low frequency part (LF) and learn the dictionary for high frequency part for further reconstruction purposes. After observing high frequency part, we perform unsupervised clustering algorithm like affinity propagation in order to detect undesirable noise patterns without prior knowledge about the number of clusters.