Presentation is loading. Please wait.

Presentation is loading. Please wait.

Scalable Video Compression for Streaming Over Internet Using DCT coefficients Zahid Ali 05030126 cs584s05.

Similar presentations


Presentation on theme: "Scalable Video Compression for Streaming Over Internet Using DCT coefficients Zahid Ali 05030126 cs584s05."— Presentation transcript:

1 Scalable Video Compression for Streaming Over Internet Using DCT coefficients Zahid Ali 05030126 cs584s05

2 MPEG Encoding Process [4]

3 Decoding

4

5 Base layer size 6 coefficients using threshold coding 128x128 Base layer size 6 coefficients using threshold coding 256x256 Observation: Using same number of DCT coefficients for different resolutions will produce video of different qualities with video with smaller resolution of inferior quality. Smaller resolutions contain more edges in sub-blocks (8x8) thus more coefficients are required which implies small resolution video will give lower compression ratio

6 Base layer size 6 coefficients using threshold coding 128x128 Base layer size 6 coefficients using zonal coding 128x128 These plots show S-Hat quality measure for threshold coded video and zonal coded video of foreman. Threshold coding handles edges better than zonal coding therefore threshold coding has performed better than zonal coding here.

7 Base layer size 6 coefficients using threshold coding 128x128 Base layer size 6 coefficients using zonal coding 128x128 These plots show S-Hat quality measure for threshold coded video and zonal coded video sequence of akiyo. Unlike foreman sequence where zonal coding failed to perform well, here in case of akiyo sequence where foreground and background is calm and thus different image is almost black,it performs better.

8 Compression Ratio for foreman Ratio for threshold coding and Static Huffman Coding 256*256 Ratio for threshold coding and Static Huffman Coding 128*128 Frames at size 256x256 get more compression ratio because of the less edges in 8x8 sub-blocks. Edges translate to high values in frequency domain thus more high values for less resolution image and lesser compression ratio

9 Compression Ratio for foreman Ratio for threshold coding and Static Huffman Coding 128*128 Ratio for zonal coding and Static Huffman Coding 128*128 Static Huffman produces less compression because a fixed “run- length/bits-required” table is used which does not give optimal compression.

10 Compression Ratio for akiyo Ratio for zonal coding and Static Huffman Coding 128*128 Ratio for threshold coding and Static Huffman Coding 128*128

11 Compression Using Dynamic Huffman Ratio for threshold and static Huffman Coding 128x128 akiyo Ratio for threshold coding and Dynamic Huffman Coding 128x128 akiyo If Dynamic Huffman is used with zonal coding compression ratio will improve (Since in threshold coding all 64 elements are run-length coded in base layer whereas in zonal coding only first few are run-length coded in base layer ) but it will also compromise quality.

12 Foreman Results At 256x256 Foreman From Original Sequence Foreman At Base layer of size 6

13 Foreman Results At 256x256 Foreman At Base layer of size 4Foreman At Base layer of size 2

14 S-Hat for foreman 256x256 If Base Layer of size 2 is used we still get quality above 3

15 S-Hat for foreman 128x128 Original Foreman Foreman At Base 10 S-Hat with mean value 3.7085

16 Akiyo S-Hat At 256x256 Original akiyo frame Akiyo at base layer of 6S-Hat at base 6 Akiyo at base layer of 2

17 S-Hat for Akiyo 128x128 Akiyo at Base 10 S-Hat with mean of 3.51 Original Akiyo Akiyo at Base 7 S-Hat with mean of 3.25

18 News S-Hat at 256x256 News Original FrameNews Base Layer size 2 S-Hat for News Base Layer size 2 here mean quality falls below 3

19 News S-Hat at 256x256 News Base Layer Size 4 Mean quality above 3

20 News S-Hat at 128x128 Original NewsNews at Base 7S-Hat with mean value of 2.7 News at Base 11 S-Hat with mean value of 2.9

21 Conclusion Videos with more texture i.e. edges (high frequency components) require more DCT coefficients to obtain minimal video quality as shown in case of foreman and akiyo video sequence. Smaller resolution also require more DCT coefficients hence smaller compression ratio because there sub-blocks i.e. 8x8 sub-blocks will likely contain more edges. In experiments conduced it is observed that if the resolution is reduced from 1 to ¼ of the original video than to obtain the same quality measure at both resolutions, the video with smaller resolution will require 4 times more DCT coefficients in it base layer. The ratio by which the video resolution is reduced than by same ratio the number of DCT coefficients should be increased in video with smaller resolution.

22 References [1] Ref Halsall, 2001 Multimedia Communications Applications, Networks, Protocols and Standards. [2] K. Cabeen, P.Gent; Image Compression and Discrete Cosine Transform [3] I. Dalgic, F. Tobagi; Characterization of Quality and Traffic for Various Video Encoding Schemes an Various Encoder Control Schemes, Stanford University [4] R. Westwater; Real Time Video Compression, Florida Atlantic University [5] J.L. Mitchell W.B. Pennebaker, C.E.Fogg and D. J. LeGall; 1997, MPEG Video Compression Standard [6] Gonzalez & Woods ;Digital Image Processing Using Matlab [7] http://www.mathworks.com/matlabcentral/fileexchange/loadFile.do?objectId=8761 http://www.mathworks.com/matlabcentral/fileexchange/loadFile.do?objectId=8761 [8] http://www.sm.luth.se/csee/courses/sms/047/videofiles/http://www.sm.luth.se/csee/courses/sms/047/videofiles/


Download ppt "Scalable Video Compression for Streaming Over Internet Using DCT coefficients Zahid Ali 05030126 cs584s05."

Similar presentations


Ads by Google