After the video codecs were investigated, some results were produced which are shown here. The Quality was measured by means of a Mean-Opinion-Score but the Storage, Data-Rate and Compression Time were measured by a computer analysis.Basically, the original video footage was compressed several times using the various codecs of interest and then the results observed to realise the advantages and disadvantages of each.

The following bar charts show how the Data-rate and Storage Requirements vary with increased Compression: (PLEASE NOTE: The File-Size is in (Kbytes) and the Data-Rate is in (KBytes/Sec))
(Samples compressed using PII-400MMX computer)
 
 

The MPEG-4 compressor used in this analysis is actually a H.263 standard codec and is a subset of the actual MPEG-4 that is out now.
 

Another more important measure is the Quality since it shows how good or bad the visuals are, of the video being viewed. The quality varies as the compression ratio is increased and varies in different ways according to which
 

codec is used since each codec has its own ability or unique properties. Due to not having the means to measure the quality by a computer analysis etc, the video is rated by using the following key by a Mean-Opinion-Score which is
 

(5) - VERY GOOD,  (4) - GOOD,  (3) - FAIR,  (2) - POOR,  (1) - VERY POOR
 

an approximate but fair type of method. A group of 20 volunteers, who have a good knowledge of what quality is, are shown the video and then each one rates the quality using the key mentioned above. The mean value is then calculated and the quality approximated to see which codec performed the best and which performed the worst.

The volunteers are given reference quality samples which indicate what is VERY GOOD quality and what is VERY POOR, in order for them to make a decision whether the sample that they are observing is closer to the best sample or the worst. The following video samples, previously used, are used here as reference points:
 

 ORIGINAL VIDEO FILE

This is the original video file and shows VERY GOOD quality.
 
 
 
 

INTEL INDEO R3.2 - (Video Information 100%)

This is a video file by Intel Indeo R3.2 and shows VERY POOR quality.
 
 

The volunteers had the option to rate the video as either (4)-Good, (3)-Fair or (2)-Poor since the best video is obviously the Original and the worst is the Intel Indeo R3.2 compressed at 100%, shown above.
The end results can be seen below in a bar chart form:
 

Obviously, the Indeo Video 5.06 codec seems to have performed best in terms of quality which uses Wavelet Technology for compression. Now that quality has been looked at, the next measure is Compression Time which

shows how long each of the individual codecs took to compress the original video at the various compression ratios.
The 'Cinepak Codec by Radius' is plotted on a chart separately because it takes the longest time and therefore stretches the y-axis to 61 secs therefore the other time values of the remaining codecs (which have a maximum time of 12 secs) would not be seen very clearly.
 

VIDEO CODECS (Summary)

After investigating the storage, data-rate, compression time and quality issues related to the various video samples used, some points are noted which classify each individual codec in terms of their unique capabilities, in other words, what each codec is good at.
The points noted are as follows:
 

1. Cinepak Codec by Radius          (uses Vector Quantisation and Frame Differencing)
 

• Since the codec did not provide further video samples compressed at various ratios, it could not be exploited any further.
• The video was compressed to 187 KB/sec and this requires a high channel bandwidth making it less suitable for Internet applications where video is used as a means of communication or freeware.
• Compression time was 61 seconds even after some compression was applied, pretty slow compared to the other codecs which follow.
• However, the speed of playback was very good although it was not mentioned and the quality was good the reason being that very little compression must have been applied since the file size was minimised to only 934 Kbytes, the original being 945 Kbytes.

2. Intel Indeo R3.2          (uses Vector Quantisation only)
 

• The file-size could be compressed to 324 Kbytes making this codec particularly useful for those interested in distributing video over the internet to web surfers who need a faster download which does not require too much storage space. Useful also for those who need to display video on their web pages but have file-size limitations.
• For 100% compression, the video showed a minimum data-rate of 65 KB/sec making it suitable for internet transmission using ISDN for instance as a means of transmission.
• However, the quality became pretty poor at 40% compression, degrading further as the compression ratio increased. At 100% it showed the worst quality out of all the various video samples which were generated.
• While providing good compression and low data-rate, the quality suffered quite considerably.
• Taking a maximum of 5 seconds to compress, this codec performed quite well compared to Indeo Video 5.06 .

3. Mpeg-4 (High Speed Compressor)         (uses Block Discrete Cosine Transform and Motion Compensation)
 

• Provided a file-size of 72 Kbytes at just 20% compression while maintaining better quality than Intel Indeo R3.2 at this stage. Although the minimum file-size achieved was the same for Intel Indeo R3.2 and Mpeg-4 as 65 KB/sec, the quality of Mpeg-4 at each stage was far more superior than Intel Indeo.
• Outstanding minimum data-rate achieved as 13 KB/sec at 100% compression while giving fair quality.
• As the compression increased, the video file began to lose some motion (very little) where only the head of the person is seen moving a little and everything else is pure still. Since the head was moving the most in the video, the Mpeg-4 codec concentrated on the moving area and distorted the surrounding image quite a lot turning white to blue etc.
• This codec showed very low data-rate at just 20% compression making it highly useful for internet applications etc.
• Performed best in terms of compression time taken, taking a maximum of 3.9 seconds.

4. Indeo Video 5.06          (uses Wavelet Compression)
 

• The file-size was compressed to a minimum of 284 Kbytes at 100% compression which is not bad, the quality being a bit below fair level. Even at 20% compression, giving very good quality, the file-size was 593 Kbytes which is okay for those who still use the 1.44 floppy disk as a means of download from the internet.
• The data-rate was not low at 20% compression (118 KB/sec) but minimised to 56 KB/sec at 100% compression. This data-rate is quite suitable for transmission and even the quality at 100% is not too bad.
• This codec uses the Wavelet Technology and provided the best quality at the various compression ratios compared to the other codecs. Up until 60%, the quality without a doubt remains on a good level where at 100% the image is distorted but most of the video information is still there and the message gets across quite clearly. According to the volunteer ratings, the quality came out best from this codec when compared to Intel Indeo R3.2 and Mpeg-4.
• However, this codec took the longest time for compression compared to Intel and Mpeg-4 taking a maximum of 12 seconds, bearing in mind that a pentium-II 400MMx machine was being used.

The advantages and disadvantages have become quite clear now. Nevertheless, bearing in mind that a fast processor was used for the compression, the compression times are still too high when compared to the new wavelet method results by Summus Technologies, mentioned in the 'New Research and Technologies' section.
 
 

Copyright 1999  by Irfan Rashid.  All Rights Reserved.