"... In PAGE 6: ... IP network and JPEG video streams. server separates the frames of a JPEG video into four layers ( Table2 ) and transmits them via four multicast addresses. It consumes 1.... ..."

"... In PAGE 13: ... To further reduce the run-time of the O-RVBR algorithm (especially for complete 90 minute videos) the acceleration schemes can be combined. Table2 shows results for Table 1: run-time for Terminator 2 trailer O-RVBR without acceleration Empirical envelope extrapolation short time noc mean estimation run-time 4h 23min 1h 34 min. 23 min.... ..."

"... In PAGE 7: ... Figure 5 and Table 1 show system construction and specifications of our evaluation environment respectively. Table2 shows specifications of a video sequence and its example images with various quality are shown in Figure 6. Table 1.... In PAGE 9: ... Still images with resolution and quality higher than the moving pictures are necessary for very detail observation and evidence. The video sequence in Table2 is encoded with multiple quality layers and resolution levels to satisfy various monitoring purposes and one bit is allocated to the highest quality layer to observe faces and small things in detail in 1280x960. Table 3 shows subjective quality evaluation of moving pictures.... In PAGE 12: ...espectively. The quality layer of ROI is three. The bit rate, decoding time and drawing time decrease as the total resolution of ROIs and an overall image decreases while the transcoding time is almost constant as in Figures 7 and 9. OLRH provides fine visual effectiveness and efficiency described in Section 3 by setting ROIs with quality = 3 and no down-sampling and an overall image with quality = 1 and 1/16 down-sampling or quality =0 and 1/64 down-sampling for the video sequence in Table2 . An overall scene with 1/4 down-sampling disturbs an observation because of large eye movement.... ..."

"... In PAGE 19: ... We analysed an SRU/SR combination with one bu er per SRU. The resulting CLR for the rst three inputs, to which video sources are connected, is given in Table1 . The precision reached with 20000 iterations is about 10?9.... ..."

"... In PAGE 9: ... 5.1 Video Workloads Our evaluation utilizes a suite of 12 video workloads that are de- scribed in Table1 ; some of the videos are also pictured in Fig- ure 20. The suite represents three common usage scenarios for loss- less video formats: Internet screencasts, computer animation, and digital television production.... In PAGE 11: ... Apart from the first frame of a video2, performing pixel transformations directly on compressed data will never in- crease the size of the file. This is illustrated in the middle columns of Table1 8, in which the output sizes are mostly equal to the input sizes (up to 2 decimal places). The only exception is contrast ad- justment on anim-scene1, in which the output is 2% smaller than the input due to variations in the first frame; for the same reason, some cases experience a 0.... ..."

"... In PAGE 4: ... Note that our formulation covers all possible scenar- ios. Table1 summarizes all possible combinations, where a2 denotes the video length, a3 and a0 are the lengths of server and proxy prefix, respectively. 3 Problem Formulation Having defined an architecture for proxy assisted video de- livery we proceed now to define the problem of selecting the proxy prefix size for each video in each community and the server prefix size for each video.... ..."

"... In PAGE 5: ...Table1 . For each combination of parameters, we conducted multiple simulation runs and computed the 95% confidence intervals of the expected number of blocks accessed and the expected service time of the most heav- ily loaded disk.... ..."

"... In PAGE 5: ...Table1 . For each combination of parameters, we conducted multiple simulation runs and computed the 95% confidence intervals of the expected number of blocks accessed and the expected service time of the most heav- ily loaded disk.... ..."