Error concealment in encoded images and video
Abstract
When transmitting compressed video over a data network, one has to deal with how channel errors affect the decoding process. This is particularly problematic with data loss or erasures. In this thesis we describe techniques to address this problem in the context of networks where channel errors or congestion can result in the loss of entire macroblocks when MPEG video is transmitted. We propose a technique for packing compressed data into packets, with the aim of detecting the location of missing macroblocks in the encoded video stream. This technique also permits proper decoding of correctly received macroblocks, and thus prevents the loss of packets from affecting the decoding process. We then describe spatial and temporal techniques for the recovery of lost macroblocks. Spatial restoration is performed by modeling the image as a Markov Random Field (MRF) and then obtaining the maximum a posteriori estimate of the missing data. We also describe a technique that can be implemented in real-time. In temporal reconstruction, we classify the available motion vectors into 9 classes via a ternary tree. Each class is assigned a cost and the vectors belonging to the class with minimum cost are modeled as an MRF. The map estimate of the motion vector belonging to the class having the smallest cost is obtained. If there is more than one map estimate for a missing motion vector then the vector that “best” preserves macroblock boundaries is chosen. We also propose the use of unequal error protection for protecting images coded by means of rate scalable algorithms, such as the Embedded Zerotree Wavelet algorithm. More important data such as IZ (isolated zero) coefficients and ZTR (zerotree) coefficients are provided with stronger protection as compared to the NEG (negative significant) and POS (positive significant) coefficients. The coded stream is then interleaved prior to transmission.
Degree
Ph.D.
Advisors
Shroff, Purdue University.
Subject Area
Electrical engineering|Computer science
Off-Campus Purdue Users:
To access this dissertation, please log in to our
proxy server.