Compressed Image Quality Measurement

Abstract

The strict requirement of the Nyquist criterion imposes acquiring large amount of data. These data when converted to compressed domain can be represented by very few data points. Due to which most of the samples are ignored. So in any signal processing system efficient use of the sensors, the memory requirements and the computational cost are not optimum. This give rise to increase in power requirements, computational complexity and over use of memory storage, which indirectly increases the cost of the system. Generally the data is stored in compressed domain to reduce the memory requirements. The calculation of the compressed coefficients requires processing time, which is dependent on the number of samples acquired. In most of the Digital systems there is only requirement of estimation of parameter of signal. These parameters are generally computed in the spatial or time domain, which again requires calculation of the inverse of the compressed coefficient. Instead if we were to calculate the parameter in compressed domain itself then the time for inverse conversion would be avoided. To further reduce the time and storage requirement one can make use of CM theory. The theory states that the compressed samples acquired can be used for certain parameter estimation. It also helps in reducing number of computations required, with less error in estimation. One of such parameter to be estimated can be the quality of an image. Quality estimation is required to provide an objective score to an image. SSIM is one of the quality score under consideration of this thesis. The implementation of compressive measurement with SSIM is the main objective of this thesis. This incorporation will help in reducing the computation which will help in developing a real time system for estimation of quality for stream of data like HD video streaming. The thesis provides with statistical results in support of the developed quality estimation metric.