IMAGE COMPRESSION USING COMPRESSED SENSING AND VECTOR QUANTIZATION

Abstract

With recent advances in the field of signal acquisition in the form of compressed sens-ing (CS) it becomes absolutely inevitable to look for its integration in previously existing fields. Image compression seems to be an absolute fit for integration not just because the word compression directly welcomes researchers to use compressed sensing but there arc a plethora of high spectral images applications that could use reduction in amount of data acquired. For example medical imaging, remote surveillance and spectroscopy etc. But before any practical applications in image compression using CS could come up, there should be sufficient empirical proof and practical algorithm available for CS based image compression that could meet the performance of current standards in image com-pression. A few algorithms were proposed in the past few years but they are not able to meet the performance of any of the standards in any scenario. We proposed a vector quan-tization and CS based image compression algorithm that combines multiple sparse vectors to achieve very high compression and can give performance similar to SPIHT standard in image compression. We utilized combining of multiple structured sparse signals to achieve compression in our proposed algorithm. But, can we combine and recover a general group of non-structured sparse signals? This question inspired us to introduce the problem of recovery of multiple sparse signals from single measurement vector. Here, measurements(all of same dimension) from multiple sparse signals are taken with different measurement matrices and added to-gether to give single measurement vector. Motivated by the data separation field existing in CS literature where dictionaries are adapted to separate components of multim.odal data, 'we modified two CS Recovery Algorithms to solve the problem of multiple sparse signals recovery from single measurement vector and presented experimental results for verifying recovery.