Implementation of sparse representation and compression distance for finding image similarity

Now a days the work on Sparse representation of signals has emerged as a major research part. It is well-known that the many natural signals such as image, music and video signals are represented sparsely if decomposed by using a proper choosen dictionaries for e.g. formed of wavelets bases. Sparse representation and compression distance is the representation that account for most or all the information of signal with linear combination of small number of elementary signal or atoms. A dictionary can be over complete or complete depending on the number of bases it contains is the same or greater than the dimensionality of the given image or signal. Traditionally, the use of predefined dictionaries has been common in sparse analysis. However, a more generalized approach is to learn the dictionary from the signal itself. Learnt dictionaries are known to outperform predefined dictionaries in several applications. In order to train a dictionary a large number of patches need to be extracted. For dictionary learning we used a K-SVD algorithm.