Ending the Debate on Clinton's Darkened Ad

[berni_mccoy]

As we all know, there has been quite a bit of debate about the Ad "True" published by the Clinton campaign. There are claims that she darkened and widened Obama's face and the video comparisons have been all over the web. Fact Check decided to clarify the issue with this article: http://www.factcheck.org/elections-2008/did_clinton_darken_obamas_skin.html

Their conclusion: "The Obama frames from the ad do appear darker than other video of Obama from the same event" but "without further evidence to the contrary, we see no reason to conclude that this is anything more than a standard attempt to make an attack ad appear sinister"

However, no one, including Fact Check, has yet done a true analysis of the digital sources of the videos and published the results. Furthermore, no one has compared other Clinton ads that contain Obama's face in them with this one. This is why this painful debate continues. Further clouding the issue are bloggers posting a variety of YouTube videos from various chains of sources, including the ObamaGirl's videos. All of these comparisons ignore the technical details required to make a fair comparison or with the context that Clinton's ad team are professional graphical artists and marketing professionals. They know what they are doing.

My previous posts on this have been less about the racial implications of any particular conclusion and more about the accuracy of any single comparison. Even Fact Check failed to perform the proper comparison: source-to-source, rendered on the same device.

Well, I'm here to put an end to the nonsense. The goal of this article is to perform a thorough analysis of the digital sources, MSNBC's online digital feed and Clinton's hi-quality online version of her ads. I think you will find the results interesting. While I will draw technical conclusions, I will not draw conclusions about intent. I do however, reserve the right to raise questions.

Before I begin, I will state with full disclosure that I am an Obama supporter. However, I am laying out the steps of this analysis so anyone can reproduce them. I encourage others to reproduce this data, as I have done, so they can convince themselves that it is accurate and they may draw their own conclusions.

I begin with establishing the base-line of the analysis: the only fair analysis is to compare the digital sources against each other as they are displayed on the same device at the same time. Any other comparison is not valid because transmission, rendering and display devices may all vary. For my own analysis, I performed all rendering of video using QuickTime 7.4.1.14 on a Mac Book Pro 2.33 GHz with 2 GB running Mac OS X 10.5.2.

Step 1. Obtain the direct source video for both the MSNBC debate and the Clinton ad "True". Clinton's hi-quality source is available here: http://www.msnbc.msn.com/id/21134540/vp/23361919#23361919 The MSNBC video does not provide a source link, although it is a high-quality feed at a resolution of 448x336. Note, that this is the same aspect ration (Width over Height) as the Clinton ad video, so scaling one to the other will not introduce stretching of video. Also, the MSNBC video had to be captured using a video capture tool as it was rendered. I used IShowU version 1.44. There is no significant difference between the captured video and the debate video as it is streamed from MSNBC.

Step 2. Render both videos, side-by-side at the same time on the same device. This was straight forward and yielded the results that many have been claiming and that have been confirmed by sources like Fact Check: the Clinton ad is darker. Because the Clinton video is at a slightly higher (30%) resolution, it was downscaled to match that of the MSNBC video. This did not change the color or aspect ratio of the video in any way. Here is a screen capture of the side-by-side rendering of Obama at the same time in the debate:



Step 3. Compare color and brightness of same area of face.  I used the Digital Color Meter (DCM) on the Mac to sample the same region of the face in both videos.  In the image below, the blue rectangle indicates the region sampled with the DCM tool. This patch was chosen due to the easily identifiable location.



In the MSNBC video, this region averages an RGB value of R:149, G:88, B:65, where each value ranges from 0, completely void of color to 255, completely saturated with color.  The average of all three numbers is an indicator of brightness, 0 being black, 255 being white (there are several ways to calculate this, but a simple technique is to take the average of all three colors).  The average luminosity therefore of the MSNBC digital feed is 101.  In Clinton's ad, we have R:104, G:87, B:70, averaging to 87.  That's about a 10-15% darker in the Clinton ad than in the raw digital source from MSNBC. Typical luminance conversions would have the difference being larger since blue is the least visible color to the human eye, but the average is good enough determine that the Clinton ad is measurably darker.

Step 4. It is obvious to the naked eye that the aspect ratio of the debate video in Clinton's ad has been stretched horizontally.  This does not happen as a by-product of compression or other image transmission, though it could happen as a result of rendering. This is why I render both videos at the same resolution and same aspect ratio, the smaller of the two (448x336). Given that we are comparing both sources, the scaling of the debate video in Clinton's ad had to have been performed intentionally. However, what the intent was, can not be determined from the scale difference alone. For example, they may have been trying to fit Obama's face in the square allotted for his face in the ad. The problem is, doing so did not require changing the aspect ratio resulting in a wider face.

Step 5. Examine Obama's face in other Clinton ads.  I downloaded the Clinton ad called "Debate" available here: http://static.hillaryclinton.com/i/video/vids/ad_20080213_debate.zip.  It is also a QuickTime movie file at a resolution of 640x480, just like the other Ad video. Here is a screenshot from that video: 

Sampling the same region of Obama's face, I performed the same color analysis from step 3 above, only shrinking the sample region by 50% since his face is 50% smaller in this video.  The sample here is R:166, G:97, B:47, which averages 103.  This is nearly an identical result to the MSNBC color, which is amazing given that the picture of Obama is not from the debate video. Also note that the aspect ratio of his face in this video is not changed and that it is roughly 50% the size of his face in the MSNBC video with the same proportion.

Technical conclusions: the source video of Clinton's ad is measurably darker than the source video of the MSNBC debate video. It is also measurably widened from the debate video. The darkening and widening are not by-products of any compression, transmission or rendering operations as none of that took place in this comparison of the two. Therefore, these modifications to the MSNBC debate video were intentional.

But what was the intent?

Why was only the Clinton "True" ad, an attack ad on Obama darkened and stretched, while the Clinton "Debate" ad, also an attack ad on Obama, not darkened or stretched?

Could it be the timing of the release of the video? The "Debate" ad was released before the Iowa caucus while the "True" ad was released prior to the Ohio primary. Did the fact that 20% of the Ohio voting population said that race was an important factor<1> in their decision have anything to do with the modification?

These are serious questions of intent that should be answered. To date, Clinton's only response on this ad, aired on Fox News<2>, were that the differences were due to image processing that occurs during compression, transmission and rendering. Clearly, this is not a reasonable explanation.

<1> source: CBS 60 minutes.
<2> source: Fox News as captured on YouTube: http://www.youtube.com/watch?v=VipVmVzwmcU