Here is a comparison of Dr. Costella's edge line and gap measurements in (a) with our drum scan gap measurements in (b):
Our new drum scan gap measurement is 3.4 times larger than Costella's findings. The primary reason for this difference is that Dr. Costella assumed the window edge and the pedestal edge are parallel to each other when in fact they are not. He rotated his image so that a vertical edge of the pergola structure was vertical in the image, and assumed that made both the window and pedestal edges horizontal. Because the image is clearer near the right side of the pedestal, he based his edge line position on data primarily from that location.
The whole point of Dr. Costella's line-of-sight argument was to measure the vertical distance between the lower right window corner and the upper left pedestal corner. Ironically, the combination of his assumptions and methods result in a gap measurement that has nothing to do with the left side of the pedestal, and hence, nothing to do with the true line-of-sight. Dr. Costella measurement of the vertical distance between the lower right window corner and the upper right pedestal edge led to his faulty conclusion.
The drum scan gap measurement properly positions the edge lines along the window and pedestal edges. We made no assumptions about the orientation of each edge. The distance between the two is the local vertical distance between the lower right window corner and a line extended leftward from the pedestal surface. The local vertical is the orientation of vertical features in the surrounding pergola structure. If you stood in Dealey Plaza where Mary Moorman did and bent your knees to align the window and pedestal edges, your eyes would move down the local vertical.
Here is a 3D intensity graph comparison of the two gap measurements. The "Costella gap" graph was made from Dr. Costella's "zoomed" version of the Moorman image, whereas the "Drum scan gap" graph was made from the drum scan version. The images are rotated 90 degrees clockwise in the graphs. The razor-thin "sheets" represent the measured edge lines. These are supposed to intersect the pedestal and window edges about midway up each slope.
Notice that in the drum scan graph on the right, the line representing the pedestal edge properly runs parallel to the surface of the pedestal. Dr. Costella's line intersects the pedestal surface only on the right side (closest to us in the graph). The window edge line is also more parallel to the window edge in the drum scan measurement.
And most importantly, the gap in the drum scan version actually covers the distance from the window edge to the pedestal edge.
In a blurred edge, the position of the original edge falls between the light and dark regions surrounding the edge. To locate an edge, Dr. Costella measures the intensity (brightness) of the light and dark regions and calculates their average. He then looks through the blurred area for the closest match in intensity to the calculated average. That is the position of the original edge.
The graph on the right shows the areas one should select above and below points from, marked with the letters A and B. The calculated average is marked midway between the two. At this level, one finds the correct width of the gap, in this case 50 pixels. However, if the average is not correct, meaning it is higher or lower than it should be, then the gap size will be mistated as wider or narrower respectively.
The graph below is the same type of graph, but from Costella's version of the Moorman photo. It plots the pixel intensities from the right portion of the pedestal edge, the area he measured and reported in his "Pedestal undamaged" data set. That area is one of three he measured along the pedestal edge. The letters A and B mark the areas from which he selected his above and below points.
Comparing this graph to the prior one, one can see that the below points, marked with a B, are too low. He should have selected points from the flatter area up around the location of the X. As a result, the calculated average is lower than it should be, the edge line is higher above the pedestal than it should be, and the gap is narrower than it should be.
Two of his three data sets for the pedestal exhibit the same problem. All of his pedestal above points are correct, as are all of his window points.
Below is the graph of the corresponding selection points from the right portion of the pedestal for the new drum scan gap. Both the above and below points are from the flatter parts of the curve surrounding the edge as they should be, making the drum scan pedestal edge line location more accurate than the Costella one.