The admissibiity of MPEG encoded video

They work by accepting “periodically chosen frames to be compressed ”¦, predicts the content of intervening frames, and encodes only the difference between the actual and the prediction. Audio is compressed synchronously. The encoder includes a decoder section in order to generate and verify the predictions. At the display, a much simpler decoder becomes possible.”

The most recent version, known as MPEG-4, is a more “universal and efficient coding for different forms of audio-visual data, called audio-visual objects.”

Some have questioned the validity of the MPEG-4 encoding relating to visual evidence. Why? Because image and video compression involves encoding a digital image or sequence of images using as few bits as possible while maintaining its visual appearance.  This results in a trade off between the image quality requirements and other needs of the application such as the target bit rate (i.e., the maximum bit rate in terms of bits per second), storage capacity and recording duration.

    Grant Fredericks, a well-known, well-respected video evidence expert, in a response to my email on this topic, had the following comments:

“”¦ cases concerning the issue of inadmissibility of MPEG video are based on myth, not reality.  Although experts have a number of legitimate concerns about MPEG (either MPEG 1, 2, 4, and in all their levels and forms), no one can state that MPEG is an unreliable format.

Forensic Video Experts are more concerned with issues relating to transcoding original DVR video to a compressed format than with the fact that the original video is compressed.  LEVA’s document “Best Practices for the Acquisition of Digital Multimedia Evidence” describes in detail the dangers of compressing already compressed evidence.

Most of our problems stem from digital video recorders (DVRs) that compress original proprietary data into an MPEG format for convenience.  We advocate acquiring the data that was produced ‘in the first instance’; if that data is in the form of an MPEG data-set, that is fine.  What we don’t want is a DVR that encodes compressed video data and then compresses that data again upon output.

Motion Prediction and LongGOPs can create problems, but as pointed out, if an expert can accurately interpret the images and the images are properly entered, there shouldn’t be a problem.  I think the concerns increase when the person entering the evidence is not qualified to assist the court with an accurate interpretation.

For example, I had a case last year in Michigan that centered around MPEG-2 video.  Due to the level of the MPEG-2 video data, the system was unable to calculate the predictive position of a person in motion on the video.  The result was that a P-Frame was calculated that placed the person in a previous position on a subsequent image.  The inaccurate placement of the person was specific to a charge of homicide against a police officer.  

At trial, I provided the court with a pixel calculation identifying the exact values of the previous Bi-Directional Frame.  It was identical to the Predictive Frame.  The jury understood that the positioning of the person (incorrectly showing the person stepping backward one step) was wrong.  The officer was acquitted.

I was asked to examine a video where the officer stated he examined video from the DVR showing the suspect turned toward a building.  However, the officer had captured the video directly from the DVR to a DVD burner using MPEG-2.  The DVR was never seized.  Upon viewing the video prior to trial, the officer noted that the image of the suspect turning toward the building is no longer there.  Instead, the last image (pixels) representing of the suspect in motion is frozen for 26 seconds.  The suspect disappears when a vehicle drives by.  The image that showed the suspect turning, was not recorded within the 15 frame GOP because of the ‘level’ of the MPEG encoding … the pixels representing the area where the suspect was in motion, did not meet the sensitivity threshold of the encoder while it calculated the GOP.

As long as the expert can articulate to the court the admissibility questions, he’ll be okay.”

I agree with Grant Fredericks. Canadian courts have applied five criteria when determining the admissibility of visual evidence: (1) relevance, (2) truth and accuracy; (3) fairness (absence of intention to mislead); verification by a sworn witness (authentication); and probative value (must outweigh prejudicial effect).  If the authenticating witness is an expert, then he or she must first be qualified as an expert before being permitted to testify.

Video compression, it could be argued, may have an affect on criteria (2) and (3).  

To the best of my knowledge, information and belief, there is no reported Canadian case in which a video was inadmissible because it was compressed.  I welcome your comments.

Elliott Goldstein can be reached at egold@rogers.com