文档介绍：Error Resilience and Error Concealment in SPIHT
指導教授：楊士萱
報告者：鄭博峰
日期：10月9日
Outline
Wavelet Transform
SPIHT Coilter and it’s computation is more complex than 5/3 filter.
9/7 filter gather more energy in lowest subband than 5/3 filter.
Energy distribution(9/7-F,”Lena”)
Lifting scheme, 5 level decomposition, level5-level3
Energy distribution(5/3,”Lena”)
Lifting scheme, 5 level decomposition, level5-level3
Outline
Wavelet Transform
SPIHT Coding
Multi-description in SPIHT
Channel
Error Concealment
Simulation
Future work
SPIHT quantization
LSP：important coefficient (grater than
threshold)
LIP ：coefficients which unknow their
importance.
LIS ：coefficients which are type A or type
B.
Introduction SPIHT
Node test
Type A
(兒子點以後)
Type B
(孫子點以後)
有重要 ：對兒子每個點做點測
試
都不重要：形成零樹後面不需再
做測試
有重要 ：將兒子兒子變成
Type A
都不重要：形成零樹後面不需再做
測試
重要
不重要
子孫重要是與否
重要
不重要
形成零樹
Outline
Wavelet Transform
SPIHT Coding
Error resilience: Multi-description in SPIHT
Channel
Error Concealment
Simulation
Future
Multi-description in SPIHT
We assumed that our network situation is packet loss. And this will cause serious degradation.
After doing Wavelet transform , almost energy will gather in low frequency. So we must protect the subband with low frequency.
My method is using MDC to overcome this problem.
MDC
Our MDC method is copying low frequency coefficients to adjacent coefficients. By doing this if the packet loss attack the SPIHT, we can use these copy coefficients to recovery lost coefficients.
So we must transmit redundancy bits to achieve this method.
Outline
Wavelet Transform
SPIHT Coding
Multi-description in SPIHT
Channel
Error Concealment
Simulation
Future work
Channel
We assume our channel error is packet loss,
and our image is 512*512.
When image