A flexible hardware architecture for 2-D discrete wavelet transform: design and FPGA implementation

The Discrete Wavelet Transform (DWT) is a powerful signal processing tool that has recently gained widespread acceptance in the field of digital image processing. The multiresolution analysis provided by the DWT addresses the shortcomings of the Fourier Transform and its derivatives. The DWT has proven useful in the area of image compression where it replaces the Discrete Cosine Transform (DCT) in new JPEG2000 and MPEG4 image and video compression standards. The Cohen-Daubechies-Feauveau (CDF) 5/3 and CDF 9/7 DWTs are used for reversible lossless and irreversible lossy compression encoders in the JPEG2000 standard respectively. The design and implementation of a flexible hardware architecture for the 2-D DWT is presented in this thesis. This architecture can be configured to perform both the forward and inverse DWT for any DWTfamily, using fixed-point arithmetic and no auxiliary memory. The Lifting Scheme method is used to perform the DWT instead of the less efficient convolution-based methods. The DWT core is modeled using MATLAB and highly parameterized VHDL. The VHDL model is synthesized to a Xilinx FPGA to prove hardware functionality. The CDF 5/3 and CDF 9/7 versions of the DWT are both modeled and used as comparisons throughout this thesis. The DWT core is used in conjunction with a very simple image denoising module to demonstrate the potential of the DWT core to perform image processing techniques. The CDF 5/3 hardware produces identical results to its theoretical MATLAB model. The fixed point CDF 9/7 deviates very slightly from its floating-point MATLAB model with a ~59dB PSNR deviation for nine levels of DWT decomposition. The execution time for performing both DWTs is nearly identical at -14 clock cycles per image pixel for one level of DWT decomposition. The hardware area generated for the CDF 5/3 is -16,000 gates using only 5% of the Xilinx FPGA hardware area, 2.185 MHz maximum clock speed and 24 mW power consumption. The simple wavelet image denoising techniques resulted in cleaned images up to -27 PSNR