Download presentation
Presentation is loading. Please wait.
Published bySharyl Lawson Modified over 8 years ago
1
SIMD Implementation of Discrete Wavelet Transform Jake Adriaens Diana Palsetia
2
Outline Motivation Goal 2D Discrete Wavelet Transform Daubechies Lifting Approach Conclusion
3
Motivation Increasing focus on multimedia has lead new image coding called JPEG-2000 JPEG-2000 over JPEG Achieves higher compression rate Computationally more intensive Replaces low-complexity and memory efficient block DCT with Discrete Wavelet Transform
4
Goal Improve computation of DWT Transform Lowering memory access Align memory Apply loop transformation techniques Extracting Parallelism Compute independent data in parallel
5
2D Discrete Wavelet Transform Subband Decomposition of 1-D signal 2-D DWT 1-D DWT on each row followed by 1-D DWT on each column
6
DWT Schemes Daubechies wavelet function is passed x samples to calculate wavelet coefficient require a temporary array hence not memory efficient
7
DWT Schemes (continued) Lifting Scheme memory efficient compared to Daubechies Use correlation in data to remove the redundancy Original 1-D sequence is split in even and odd indexed sequence Values are iteratively modified by predict and update step s: update step d: predict step P: predict weights U: update weights
8
JPEG Codec Analysis using JasPer Table 1: RunTime using GNU profiler for 1792x1200 bitmap image
9
Approach Initial: modify JPEG-2000 to incorporate SIMD implementation using SSE2 Current: Implement C based DWT algorithm (Daubechies 4 and Daubechies 4 with lifting) Take the original algorithm and apply subword parallel functions using SSE2 instruction set Compare Speedup of original algorithm with SSE2 implementation
10
Conclusion DWT Superior to DCT (multi-resolution analysis) Computationally complex Implement Wavelet Transform Schemes Use SIMD instruction for optimization Compare Performance
11
References Daubechies D4 Wavelet Transform http://www.bearcave.com/software/java/wavelets/da ubechies/index.html http://www.bearcave.com/software/java/wavelets/da ubechies/index.html M. Rabbani, and R. Joshi, “An overview of JPEG 2000 still image compression standard”, Signal Processing: Image Communication, vol. 17, pp3-48, 2002 A. Shabahrami, B. Juurlink, S. Vassiliads, “Performance Comparison of SIMD Implementations of Discrete Wavelet Transform”,
Similar presentations
© 2024 SlidePlayer.com. Inc.
All rights reserved.