Compressive sensing based image processing and energy-efficient hardware implementation with application to MRI and JPG 2000

Compressive sensing based image processing and energy-efficient hardware implementation with application to MRI and JPG 2000

In the present age of technology, the buzzwords are low-power, energy-efficient and compact systems. This directly leads to the date processing and hardware techniques employed in the core of these devices. One of the most power-hungry and space-consuming schemes is that of image/video processing, due to its high quality requirements. In current design methodologies, a point has nearly been reached in which physical and physiological effects limit the ability to just encode data faster. These limits have led to research into methods to reduce the amount of acquired data without degrading image quality and increasing the energy consumption.

Compressive sensing (CS) has emerged as an efficient signal compression and recovery technique, which can be used to efficiently reduce the data acquisition and processing. It exploits the sparsity of a signal in a transform domain to perform sampling and stable recovery. This is an alternative paradigm to conventional data processing and is robust in nature. Unlike the conventional methods, CS provides an information capturing paradigm with both sampling and compression. It permits signals to be sampled below the Nyquist rate, and still allowing optimal reconstruction of the signal. The required measurements are far less than those of conventional methods, and the process is non-adaptive, making the sampling process faster and universal.

In this thesis, CS methods are applied to magnetic resonance imaging (MRI) and JPEG 2000, which are popularly used imaging techniques in clinical applications and image compression, respectively. Over the years, MRI has improved dramatically in both imaging quality and speed. This has further revolutionized the field of diagnostic medicine. However, imaging speed, which is essential to many MRI applications still remains a major challenge. The specific challenge addressed in this work is the use of non-Fourier based complex measurement-based data acquisition. This method provides the possibility of reconstructing high quality MRI data with minimal measurements, due to the high incoherence between the two chosen matrices. Similarly, JPEG2000, though providing a high compression, can be further improved upon by using compressive sampling. In addition, the image quality is also improved. Moreover, having a optimized JPEG 2000 architecture reduces the overall processing, and a faster computation when combined with CS.

Considering the requirements, this thesis is presented in two parts. In the first part: (1) A complex Hadamard matrix (CHM) based 2D and 3D MRI data acquisition with recovery using a greedy algorithm is proposed. The CHM measurement matrix is shown to satisfy the necessary condition for CS, known as restricted isometry property (RIP). The sparse recovery is done using compressive sampling matching pursuit (CoSaMP); (2) An optimized matrix and modified CoSaMP is presented, which enhances the MRI performance when compared with the conventional sampling; (3) An energy-efficient, cost-efficient hardware design based on field programmable gate array (FPGA) is proposed, to provide a platform for low-cost MRI processing hardware. At every stage, the design is proven to be superior with other commonly used MRI-CS methods and is comparable with the conventional MRI sampling.

In the second part, CS techniques are applied to image processing and is combined with JPEG 2000 coder. While CS can reduce the encoding time, the effect on the overall JPEG 2000 encoder is not very significant due to some complex JPEG 2000 algorithms. One problem encountered is the big-level operations in JPEG 2000 arithmetic encoding (AE), which is completely based on bit-level operations. In this work, this problem is tackled by proposing a two-symbol AE with an efficient FPGA based hardware design. Furthermore, this design is energy-efficient, fast and has lower complexity when compared to conventional JPEG 2000 encoding.