44 2. FIVE STORIES TO A MODEL OF VIDEO STRUCTURE
would not be appropriate for detecting alternation or detecting the centers within the sequence as
identied by Bellour.
Our ultimate goal in analyzing the structure of the Bodega Bay sequence was to nd the
elements of the physical structure of the moving image document that prompted Bellour to make
the statements (tacts) he did about the lm. To accomplish this task, it was necessary to look at
the structure of the segment on at least two levels. First, Bellour breaks the sequence into “shots”
or frame sets and selects key frames. is requires an examination of individual frames. Second,
Bellour describes alternation between the frame sets, the unique character of the “hinge,” the two
centers and the gull strike. ese tacts are descriptions of the relationship between frame sets.
We sought precise, repeatable, numeric, and graphical representations of the signal that would
enable discussion of lmic structure—the message, in the terms of Shannon and Weaver. We sought
the means by which we might discuss message structure, so that discussions of meaning would have a
signicant touchstone. It might be said that we sought a method of ngerprinting the frames.
In standard digital images each and every color is composed of a certain amount of red, a
certain amount of green and a certain amount of blue—with black being the absence of any red,
green or blue and with white being maximum of each. In the frame images we captured there is a
possibility of 256 shades of red, 256 shades of green, and 256 shades of blue for a possible palette
of over 16 million colors. Deriving a histogram of each of the RGB components or the aggregated
values distributed across an Xaxis of 255 points (the zero origin being the 256th) yields a nger-
print—a color distribution map—of each frame.
Perhaps one of the most appealing aspects of mapping color distribution is that it is an en-
tirely softwarebased process. ere is no necessity for human intervention to determine and mark
what is to be considered the “subject” or how many pixels (what percentage of the frame area) make
up some viewerselected object. Not that these are not useful for some sorts of analysis, but using
just the color palette enables an essentially judgmentfree analytic process.
2.4.9 METHOD OF PIXEL-LEVEL ANALYSIS
Structural Analysis. We converted the Bodega Bay to an AVI le and then extracted the indi-
vidual frames to 12,803 JPG image les. We generated RGB histograms for each of the 12,803
frames using the Python Imaging Library. A Lorenz transformation was then performed on each
histogram. We calculated a Gini coecient for each frame to generate a scalar value representing
the color distribution of each frame. e Gini coecient compares a perfectly even distribution
of RGB against the actual distribution in each frame. We used the dierences in Gini coecients
between successive frames as a measure of change across frames.
Codifying Bellour’s Analysis. Bellour’s analysis does not include precise times or frame
numbers to either select key frames or delineate frame sets; however, he includes photographs of