Efficient Hardware Implementation Of Haar Wavelet Transform With Line-Based And Dual-Scan Image Memory Accesses
Image compression is of great importance in multimedia systems and applications because it drastically reduces bandwidth requirements for transmission and memory requirements for storage. An image compression algorithm JPEG2000 isbased on Discrete Wavelet Transform. In the hardware implementation o...
Saved in:
Main Author: | |
---|---|
Format: | Thesis |
Language: | English |
Published: |
2017
|
Subjects: | |
Online Access: | http://eprints.usm.my/39413/1/Laila_Ahmed_Saad_24_Pages.pdf |
Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
Summary: | Image compression is of great importance in multimedia systems and applications because it drastically reduces bandwidth requirements for transmission and
memory requirements for storage. An image compression algorithm JPEG2000 isbased on Discrete Wavelet Transform. In the hardware implementation of DiscreteWavelet
Transform (DWT) and inverse DiscreteWavelet Transform (IDWT),the main problems are storage memory, internal processing buffer, and the limitation of the FPGA resources. Based on non-separable 2-D DWT, the method
used to access the image memory has a direct impact on the internal buffer size,the power consumption and, the transformation speed. The need for internal buffer reduces the image memory access time. The main objectives of this thesis are as follows; to implement a 2-D Haar wavelet transform for large gray-scale image, to reduce the number of image memory access by implementing the 2-
D Haar wavelet transform with a suitable combination between using external memory and internal memory, and targeting a low-power and high-speed architecture
based on multi-levels non-separable discrete Haar wavelet transform. In this work, the proposed two architectures reduce the number of image memory access. The line-based architecture reduces the internal buffer by 2 x 0.5 x N
where N presents the image size. This happens for the low-pass coefficients and for the high-pass coefficients. The dual-scan architecture does not use the internal
memory. Overall both architectures work well on the Altera FPGA board at frequency 100 MHz.
|
---|