This report is also available as an Acrobat file.

Contents Digital Video for Multimedia: Considerations for Capture, Use and Delivery

Section 3: Digital Video on Trial

During our investigations into digital video for multimedia applications a series of experiments were performed to investigate the various parameters involved with video capture such as, choice of codec, frame rates, key frames, compression quality and size of video window. The results, a series of digital video clips, can be accessed and downloaded from the following World Wide Web site: http://www/.ets.bris.ac.uk. Further information expanding on the issues already discussed in this report is also available, as are links to other sites with relevant information.

Method:

A video sequence containing several scenes with differing content (scenes with little change, rapidly changing scenes, scenes with multiple content and special effects, etc) was captured 'raw' and using real-time compression at various frames rates and window sizes, and compressed using a variety of different quality settings and codecs. The sequence was captured from a non-degradable video source, an analogue write once laserdisc, to eliminate tape degradation effects and maintain quality. All sequences were captured on a PC. Two cards were used in the study:

Creative Labs Videoblaster RT 300 - captures AVI and performs real-time Indeo compression
Vitec Video NT - captures real-time MPEG and AVI

Results:

Table 1: Compression times and file sizes for various quality settings and codecs using MS VidEdit

Initial file: 30seconds of Intel Raw video, 25frames per second, 320 x 240, original file size 70.4MB. All files are at 24 bit colour except MSVideo 1 as this only supports 16 bit colour.

Output settings	Codec	conversion time (minutes)	size (MB)	output file
100% quality key frame every frame	MS Video 1	9:00	70.1	msvid_1.avi
	Indeo 3.2	10:45	13	ir32_1.avi
	Indeo Interactive	17:00	12.4	ir4_1.avi
	Cinepak	48:30	18.7	cnpk_1.avi
0% quality key frame every 10 frames	MS Video 1	2:50	0.61	msvid_2.avi
	Indeo 3.2	14:15	2.65	ir32_2.avi
	Indeo Interactive	43:45	1.79	ir4_2.avi
	Cinepak	33:10	18	cnpk_2.avi
Codec defaults for quality and key frame rate	MS Video 1	3:40	5.17	msvid_3.avi
	Indeo 3.2	14:05	2.89	ir32_3.avi
	Indeo Interactive	46:55	4.95	ir4_3.avi
	Cinepak 34:50	18	cnpk_3.avi

Comparing the conversions for MS VidEdit and Adobe Premiere there was no noticeable difference in either the time taken or the final file size. Adobe Premiere does afford the opportunity to output to the Apple Quicktime format, corresponding times and file sizes are given in the table below.

Table 2: Compression times and file sizes for various codecs at the highest quality setting using Adobe Premiere

All starting with the same initial file, 30seconds of Intel Raw video, 25fps capture, 320 x 240, original file size 70.4MB

Output settings	Codec	conversion time (minutes)	size (MB)	output file
quality 5 key frame every frame	MS Video 1	6:35	62.8	video_1q.mov
	Indeo 3.2	11:40	12.6	ir32_1q.mov
	Cinepak	52:05	18.2	cnpk_1q.mov

Note: At the time of writing the Quicktime format did not support the Indeo Interactive codec. Quicktime provides a quality setting range of 0-5.

Table 3: A comparison of file sizes versus frame size.

All files captured to RAM as Indeo Raw (via the Videoblaster RT 300) and recompressed at 25 fps, highest quality, key every frame using MS VidEdit

Frame size	codec	file size (bytes)
160x120	Indeo 3.2	4420298
160x120	Indeo Interactive	4420814
192x144	Indeo 3.2	6181010
192x144	Indeo Interactive	6516170
288x216	Indeo 3.2	11681090
288x216	Indeo Interactive	11399920
320x240	Indeo 3.2	13280700
320x240	Indeo Interactive	12658408
640x480*	Indeo 3.2	1884426
640x480*	Indeo Interactive	3956890

*Note: The low file sizes here are due to the computer and card being inable to capture at 640 x 480, captures were therefore carried out manually (i.e. frame by frame) and represent sequences of 100 frames (4 seconds).

Table 4: A comparison of file sizes versus frame rate

All files captured to RAM as Indeo Raw (via the Videoblaster RT 300) and recompressed at 320 x 240, highest quality, key every frame, using MS VidEdit

Frame rate	codec	file size (bytes)
25	Indeo 3.2	13280700
25	Indeo Interactive	12658408
15	Indeo 3.2	7782486
15	Indeo Interactive	7440252
12.5	Indeo 3.2	6597990
12.5	Indeo Interactive	6331812

Table 5: A comparison of file sizes versus codec using the Video NT card.

Initial file: Approximately 17.5 seconds of Vitec 422 video (Vitec's form of raw compression), 25 frames per second, 192 x 144, original file size 25 MB, recompressed at highest quality, key frame every frame, using Vitec editing software

codec	file size (bytes)	output file
Cinepak	3792598	cnpk.avi
MS Video 1	13669916	msvid.avi
Intel Indeo 3.2	3264318	ir32.avi
Intel Interactive	3441432	ir4.avi
Vitec DCT	4484472	vitec2.avi

Note: Vitec's own DCT (Discrete Cosine Transform) codec only gave us control over some of its settings – others were greyed out.

Table 6: MPEG files.

Comparison of file size and quality settings of MPEG files created from either Indeo Raw at two different frame sizes, or compressed AVI files, using MPEG compression software (MPEG maker) supplied with the Video NT card.

Frame size	Initial codec	file size (bytes)	output file	MPEG compressor settings
192x144 file from table 3	Intel Raw	611884	hi_cmprs.mpg	default high compression
	Intel Raw	1713813	quality.mpg	default high quality compression
	Intel Raw	2474927	quick.mpg	default fast compression
320x240 original file from table 1	Intel Raw	1223695	hi_cmprs.mpg	default high compression
	Intel Raw	3250012	quality.mpg	default high quality compression
	Intel Raw	4823189	quick.mpg	default fast compression
Recompressed using files from table 1	Cinepak	4370476	ex_cnpk.mpg	default high quality compression
	MS Video	4245941	ex_msvid.mpg	default high quality compression
	Intel Indeo 3.2	3812733	ex_ir32.mpg	default high quality compression
	Intel Interactive	4073239	ex_ir4.mpg	default high quality compression

One sequence was captured directly to MPEG using the Vitec Video NT card using the 'high quality' setting but – with the stuttering playback, general poor quality and time constraints – we did not consider worthwhile continuing with this approach. The file is vitec.mpg

Initial Discussions:

The above tables reveal some interesting facts about the nature of the various codecs listed above. The nature of the relationship between quality level, compression and file size varies from one codec to the next. The balance of tradeoffs is dependent upon the video material (content and quality) and its purpose. However, assessment of the various codecs is a subjective process and the above results should be considered while viewing the files. Refer to the following World Wide Web site for further comparisons and information and to download and view the video files: http://www.ets.bris.ac.uk

Contents

Graphics Multimedia Virtual Environments Visualisation Contents