A recent California case provides yet another illustration of how conclusions premised on the analysis of images obtained from surveillance video must be reached by people who have forensic video analysis training and how it is the role of the court to deny admissibility to unreliable expert evidence. 

In State of California v. Carrington, 2018 WL 671903 (California Court of Appeal, Second District), the appellant was charged with first degree murder.  He was convicted at trial and appealed his conviction arguing, inter alia, that the trial judge erred in rejecting defense expert facial recognition evidence that would exonerate him. 

The victim had been shot repeatedly outside of a liquor store.  The liquor store had five surveillance cameras which captured key aspects of the event, including the shooting itself.  

Prior to trial, a defense expert, Dr. Paul Wallace (a facial reconstructive surgeon and facial recognition expert), examined the store’s surveillance video but found that it was less than ideal and was unable to conclusively determine whether the suspect seen in the video was in fact the appellant.  Post-conviction, a number of declarations from Dr. Wallace as well as a person who worked at a camera store were filed in support of the appellant’s motion for a new trial.  The state filed declarations in response.  Pertinent details of these declarations and motion testimony follow.

Dr. Paul Wallace

Post-trial, Dr. Wallace conducted a second evaluation of the surveillance video images using “current digital High Definition Technology” which made it possible “to tease out and enhance the images.”  He said that the “new state of the art software” was similar to facial recognition software used by law enforcement agencies and permitted him to “enhance image files for matches.”  Sixty-four facial images of the appellant were used to compare to 22 still images taken from the surveillance video.  Dr. Wallace said that “[t]he result was no match on any software program,” which caused Dr. Wallace to conclude that the suspect in the video was not the appellant.  During a hearing on the appellant’s motion, Dr. Wallace was asked how the still images obtained from the surveillance video had been enhanced to make them clearer and he conceded that he was unaware of the mechanism or technology used to enhance their quality.   

Los Angeles County Sheriff’s Deputy Manuel Cuevas

The state opposed the appellant’s motion on the basis that the new facial recognition evidence was not sufficiently reliable under the requisite Kelly/Frye analysis.  Deputy Cuevas was the project lead for Los Angeles County’s mugshot database, which was used as a facial recognition system.  He testified that facial recognition was “kind of in its infancy” and that while the technology was evolving, its reliability and accuracy were not equivalent to the reliability and accuracy of fingerprint analysis.  He further testified that law enforcement agencies did not use facial recognition as the sole basis to identify or eliminate a suspect.  

Deputy Cuevas testified that based on National Institute of Science and Technology standards as well as his own experience, between 60-120 pixels between the eyes were needed for facial recognition to be utilized.  The images used by Dr. Wallace had only seven pixels between the eyes of the suspect, an insufficient number to generate an accurate template for facial recognition.  He stated that comparing these low resolution images with high resolution photos taken of the appellant afterward was like comparing apples to oranges.  

Lindsay Mejia

Subsequent to Deputy Cuevas’s testimony, the appellant filed a declaration from Lindsay Mejia, a digital artist who worked at a camera store and who “trained in photoshop work and digital imaging for four years.”  She testified that she took photographs of the surveillance video images using a “Black Magic” high resolution camera.  In her declaration she stated that “[t]he camera does not ADD any material that was not present in the video.  Instead it enhances what is ALREADY present.  My role was to put the exhibits together.”

Preliminary Comments of the Court

During a subsequent hearing, the Court stated that it would not have allowed the initial testimony of Dr. Wallace into evidence at trial.  While Dr. Wallace is an expert in facial recognition, he is not an expert in the technology used to create the images he examined.  The Court stated that it had “…absolutely no faith in the integrity of those photos, because I heard absolutely nothing that would establish that they were reliable…”  The Court further expressed concern that Kelly/Frye was not being addressed.  The Court noted that Dr. Wallace had no expertise in the computer programs utilized and that therefore his opinion on facial recognition was lacking foundation.  The Court accordingly requested further evidence as to how Mejia created the high resolution photos.

Lindsay Mejia (reprise)

The appellant filed a further declaration from Mejia in which she stated that no pixels were added or subtracted when she created her images.  She said that the surveillance video player would not play on her computer and so she decided to “…reshoot it with a camera that can produce a file which I can view on my own computer.”

The state filed a supplemental, more detailed declaration of Mejia as to how she obtained her images of the surveillance video.  She stated that she mounted the camera on a tripod and took photos of Dr. Wallace’s laptop computer screen as it played the video.  She further stated:

…[W]hatever image can be seen on the laptop screen with your naked eye is the same thing you could look at with the recording being captured by the BM camera. …To clarify, if the original surveillance video recorded at a rate of six frames per second, the BM camera is simply recording those same six frames over and over at a rate of 24 frames per second; you cannot get more information from the surveillance video by recording it with the BM camera. …The original surveillance video is very pixilated and nothing can be done to improve that quality without manipulating the original recording. The BM camera does not increase the sharpness or the quality of the original surveillance video.

George Reis

The state filed a declaration of George Reis, an expert in forensic video analysis.  He examined the images as well as the declarations of Dr. Wallace and Mejia.  Reis stated the following:

• The surveillance video was recorded at a resolution of 352 x 240 pixels and a frame rate of seven frames per second.  By way of comparison, cell phone video typically has a resolution of 720 x 1280 pixels, resulting in 11 times more resolution.  The frame rate of standard video is 30 frames per second.  
• Dr. Wallace’s declaration that he “…refrained from using any enlargement or other type of enhancement to improve resolution” is misleading.  By photographing the video images at a higher resolution than 352 x 240 pixels, one is by definition enlarging, resampling and magnifying them.  
• Dr. Wallace’s declaration that “…we were able to obtain far more stills from the surveillance video…” is incorrect.  Many of the still images are simply duplicates of the same frames.  The camera was simply capturing the computer screen at a higher frame rate than the surveillance video portrayed.  This resulted in duplicate frames being recorded, not additional frames.
• The surveillance video captured general class characteristics only, such as approximate skin tone, gender and approximate height to weight ratio. 

Reis stated that Dr. Wallace’s conclusion that the person in the video is not the appellant is not supported by the video evidence, which Reis found to be unreliable.  He noted that the low resolution and high level of compression and significant artifacting prevents any such conclusion being possible.  Reis further observed that neither Dr. Wallace or Mejia had any training or experience in the field of forensic video analysis or any familiarity with best practices in this area of forensic science.  Reis stated that “based on the methodology used, it is evident that they are not familiar with this field.”

Dr. Wallace (reprise)

Following Reis’s declaration, the appellant filed a final declaration from Dr. Wallace in which he stated “[R]e-examining the case using the new generation MacPro with Retina allowed me to make more digital “still” images without enhancing the images. The multiple still photos have made it unequivocally certain that the person on the video is not [appellant].”  

Trial Court Ruling

The Court ruled that Dr. Wallace’s testimony was inadmissible as it did not meet the Kelly/Frye standard for reliability and was based on computer images that were so unreliable that they rendered his proposed testimony inadmissible under California Evidence Code Section 352.  The Court noted that Dr. Wallace did not understand how the images used for comparison purposes were created.  The evidence presented by Reis was preferred as being “by far, the most persuasive” information as to image unreliability.  The Court also did not accept that the “Mac computer” was new technology such that it could not have been discovered in time.  Accordingly, the Court ruled that a new trial would not be granted.

Court of Appeal Analysis and Ruling 

The appellant argued that the trial court abused its discretion in refusing to grant a new trial and that it erred in concluding that the facial recognition evidence was unreliable and inadmissible.

The Court of Appeal stated that the Kelly/Frye test renders inadmissible evidence derived from a “new scientific technique” unless the proponent establishes that the technique is generally accepted as reliable in the relevant scientific community; the witness testifying about the technique and its application is a properly qualified expert on the subject; and the person performing the test in the particular case used the correct scientific procedures.  It noted that the purpose of Kelly is to protect the jury from techniques which, though new, novel, or experimental, convey a misleading aura of certainty.  In dismissing the appeal on this point, the Court stated:

Kelly/Frye’s first requirement that a “ ‘technique is generally accepted as reliable in the relevant scientific community’ ” (Jackson, supra, 1 Cal.5th at p. 315) is usually established by expert testimony on the subject. (Stoll, supra, 49 Cal.3d at p. 1155, 265 Cal.Rptr. 111, 783 P.2d 698.) Although Wallace was a facial recognition expert, appellant failed to demonstrate that Wallace was an expert in the new technology employed to produce and enhance the still photographs derived from the low-resolution surveillance video, which photographs Wallace subsequently compared to high-resolution photographs of appellant. Wallace admitted he was “not a specialist” regarding this alleged new technology and that he was not “exactly sure of the actual mechanism” the technology used to make a photograph clearer. The trial court expressly found Wallace was not an expert in the alleged new technology. Moreover, Cuevas’s testimony and the declaration of Reis affirmatively demonstrated that the technique employed to enhance the poor-quality surveillance video was not generally accepted as reliable in the relevant scientific community, and that still photographs derived from the low-resolution, highly-compressed surveillance video could not serve as a reliable basis for comparison with high-resolution photos taken of appellant.

Appellant failed to demonstrate that the technique upon which Wallace’s testimony would be based met the first and second requirements of the Kelly/Frye test—that the scientific community accepted the technique as reliable and that the witness testifying about the technique and its application is a properly qualified expert on the matter. (Jackson, supra, 1 Cal.5th at pp. 315–316.) Further, the court did not abuse its discretion by finding, pursuant to Evidence Code section 352, that Wallace’s proposed testimony should not be admitted because the technique underlying his opinion had not been shown to be reliable. The court properly denied appellant’s disfavored new trial motion based upon Wallace’s proposed testimony.

The conviction was accordingly affirmed.

Analysis

This case illustrates three important points.  

First, only properly trained and qualified people should undertake forensic video analysis.  Unfortunately, there are many examples in the United States of people with a production video background and no forensic video training working on forensic cases, utilizing unsound processes and forming unjustifiable opinions.  There are both civil and criminal case examples of people claiming to be experts and giving evidence that would be laughable but for the solemnity of the proceedings in which they are testifying.  Though these types of “experts” are typically called by the defense, they are also sometimes called by the state.  

To be clear, having a production video background is not a bar to undertaking competent forensic video analysis.  Such a background can be invaluable in fact.  However, a production video background alone does not qualify one to undertake forensic video analysis, which is a robust scientific discipline.  A certified forensic video analyst undertakes significant academic and practical training along with rigorous proficiency testing in order to achieve certification.  These are the people who should be undertaking forensic video analysis, writing expert reports and testifying before the court as they are truly in a position to assist the trier of fact.

Second, experts who form opinions based on video images must understand all of the technical aspects of the relevant video evidence from creation to final use in court.  This is not just an ideal but effectively is a constitutional mandate where those experts are called by the state (as was not the case in Carrington).  The Sixth Amendment to the United States Constitution guarantees the right of the defendant to confront his accuser.  Accordingly, a prosecution expert who relies on video or still images to form an opinion must have a working knowledge of the technical processes that generated such evidence.  This important point was illustrated by the Supreme Court of Connecticut in State of Connecticut v. Swinton, 268 Conn. 781; 2004 Conn. LEXIS 190 (Conn. 2004).  For more information on this case, please refer to the January 2018 article on my website entitled Generated Evidence: What Must the Expert Know?

Third, an expert in one field of expertise should not offer opinions premised on the interpretation of video evidence unless he/she is also a fully trained and qualified forensic video analyst.  Expert witnesses have a professional and ethical obligation to stay within their lane and to not offer expert opinions outside of their area of expertise.  For more information on this topic, please refer to the following links to articles on my website which cover this topic and provide recent examples of experts straying from their lane only to chastised by the court for doing so.

• Case Study: Expert Witness Claims Expertise Not Actually Present (May 2018)
• Stay Within Your Lane!: The Importance of Knowing Your Boundaries as an Expert Witness (August 2018)
• Chicago Murder Prosecution Falters When Expert Fails to Stay Within His Lane (September 2018)

Dr. Wallace admittedly had no training or qualifications in forensic video analysis and did not know how the images he used for comparison purposes were created.  He should not have offered any opinions on facial recognition without the assistance of a forensic video analyst.  It is recommended (and necessary) that when more than one area of expertise is required to work on a case, that the appropriate experts be consulted and that they work together.  Had that happened in this case, it is unlikely that facial recognition evidence would have been tendered in this case as it would have been apparent that the video images were unsuitable for that purpose.

The law is well equipped to ferret out evidence that should not be placed before the trier of fact.  For that law to be utilized, either opposing counsel or the court must be aware that the evidence tendered is of questionable merit and admissibility.  In the Carrington case and the other cases cited, the impugned evidence was subjected to the proper legal tests and so no harm was done to the case before the court.  What is more troubling is the unknown number of cases where unsupportable opinions premised upon video interpretation have been accepted by the court and relied upon in deciding cases.