Practice Midterm Solutions

CMSC 427

Sample Problems (Note, some of these problems may be a bit more involved than ones
Id ask on a time-limited exam).

1. Create a matrix that rotates points 90 degrees about the point (1,1).

We can do this by translating to the origin, rotating, and translating back. We get:

1 0 1 0 1 0 1 0 1 1 0 1 0 1 1 0 1 2
0 1 1 1 0 0 0 1 1 = 0 1 1 1 0 1 = 1 0 0
0 0 1 0 0 1 0 0 1 0 0 1 0 0 1 0 0 1

We can verify that this is right by checking that it takes the point (1,1) (written (1,1,1) in
homogenous coordinates) to the point (1,1), and that it takes the point (2,1) to (1,2).

2. What is the distance from (3,2) to (7,5) in the direction (1,2)?

To do this we need to find the inner product between a vector from (3,2) to (7,5) and a
unit vector in the direction (1,2). The first vector is (4,3). The unit vector we need is
(1,2)/sqrt(5). This inner product is 10/sqrt(5)=2sqrt(5)

3. Provide any two rows of a 4x4 matrix that will transform 3D points as they would
appear in the coordinate system of a viewer centered at (3,2,0) facing in the
direction (1,2,3). The new z coordinate should describe the distance from the
viewer to a point, in the direction that they are viewing it.

We are changing the origin and the viewing direction. We can do this by combining two
matrices. If we just wanted to change origins, all we would need to do is create a 3D
translation matrix that takes (3,2,0) to the origin. This would be:

1 0 0 3
0 1 0 2
0 0 1 0
0 0 0 1

However, we also want to change the viewing direction. To do this, well create a matrix
in which the first three elements of the third row represent our new z direction. This
direction is represented by the unit vector (1,2,3)/sqrt(14). So, the matrix we want will
look like:
 1 0 0 3
0 1 0 2
1 2 3 = 1 2 3 7
0 0 0 1 0
14 14 14 14 14 14 14
0 0 0 1 0 0 0 1 0 0 0 1

Notice that to get the bottom two rows we never had to come up with the top two rows of
the first matrix on the left.

4. Suppose you have a camera with a focal point at (0,0,0) and an image plane of
z=1. Give an example of a right triangle in 3D that will also appear in the image
as a right triangle, assuming perspective projection. Do the same thing for
orthographic projection.

Perspective projection scales points by the inverse of their distance. So, if we pick points
that have the same z coordinate, they will all be scaled the same amount. So, for
example, if we take a triangle with vertices at (0,0,2) (2,0,2), (0, 2, 2), this will appear in
images with coordinates (0,0), (1,0), (0,1), and still be a right triangle.

With orthographic projection, the same triangle will appear as (0,0), (2,0), (0,2), also a
right triangle.

5. Prove that parallel lines in the world do not always appear as parallel lines with
perspective projection.

We just need an example to show this. Lets consider lines that lie in the y=-5 plane,
with our standard perspective camera. Consider a line with the equation z = mx+b. This
contains the points (0,-5,b) and (1,-5, m+b). These appear in the image as points (0, -5/b)
and (1/(m+b), -5/(m+b)). This line has a slope of (-5/(m+b) + 5/b)/(1/(m+b)) = -5 +
5(m+b)/b = 5m/b. Clearly this slope depends not just on m, the slope of the original line,
but also on b. Two lines with the same value for m and different values for b will be
parallel in the world, but will have different slopes in the image.

6. Explain how you would adapt the algorithm we learned that discretizes lines to
discretely represent the boundary of a circle. What pixels would be filled in to
represent a circle centered at (20,20) with a radius of 5?

First, lets consider the simpler, nave algorithm. We cannot just increment x from 15 to
25, and solve for two y values for each x, because in some parts of the circle y changes
more rapidly than x, so we would get a circle with gaps. Instead, we can divide the circle
into regions where either x or y is changing more rapidly. For example, at an angle
between pi/4 and 0, y is changing more rapidly. This occurs in the region of the circle
starting at (20+5/sqrt(2), 20+5/sqrt(2)) and ending at (25,20). Rounding off, this means
we must consider y values of 24, 23, 22, 21, 20. For these, the corresponding x values
are 23, 24, 25,25,25, so we fill in points (23,24), (24,23), (25,22), (25,21), (25,20).
We can try to adapt DDA to this problem by noticing that the tangent to a circle at a
boundary point is perpendicular to a vector from the center to the boundary point. So if
we start at the point (25,20), a vector from (20,20) to (25,20) is (5,0), and the tangent at
that point is 0/5. If we change y by 1, we should not change x at all, and we find the
point (25,21). At this point the slope will be -1/5, so the next point on the circle will be
(24 4/5, 22). Rounding this off, we would fill in the pixel (25,22). The slope at this point
would be -2/(4+4/5) = -5/12. So the next point on the circle will be (24 23/60, 23), which
rounds to (24,23). Continuing this way brings us to (23.7, 24), rounded to (24,24).
Results are a bit less accurate. Is this worth it?

7. Suppose you have an image with vertical, black and white stripes that are five
pixels wide, and you want to shrink this image to half its current size in each
dimension. Is there a problem with aliasing? If not, why not? If yes, how would
you prevent this by smoothing using averaging? How much would you need to
smooth? There may not be a simple, exact answer to this question, so just think
about it and describe the consequences of different possible choices.

If we shrink this image by sampling, then stripes will be represented by either two or
three pixels, depending on where we choose our sample points. This wont be as bad as
some aliasing effects, but it still will create significant artifacts. For example, suppose a
white stripe extends from 1 to 5, and a black stripe from 6 to 10, and we sample at
2,4,6,8,10. The white stripe will appear to be two pixels wide, and the black stripe will
be three pixels wide. A horizontal slice of the image will look like: (1 1 0 0 0 1 1 0 0 0 1
1 ).

If we average a great deal, the stripes will become a uniform gray. Now there will
certainly be no aliasing, although we will have completely lost the structure of the image.

Now, suppose we replace each pixel by the average of it and its neighbors. Then, after
averaging, a horizontal slice of the image will look like:

(2/3 1 1 1 2/3 1/3 0 0 0 1/3 2/3 1 1 1 2/3 1/3 0 0 0 1/3 2/3 1 1 1 )

If we sample this, we get something like: (1 1 1/3 0 1/3 1 1 1/3 0 1/3 ) This still isnt
perfect, but it will look similar to a white stripe that is 2 2/3 pixels wide, and a black
stripe that is 2 1/3 pixels wide.

We get even better results if we use weighted averaging, so that each pixel is replaced by
the average of its original value and that of half of each of its neighbors. Now, averaging
produces something like:

(3/4 1 1 1 3/4 0 0 0 1 1 1 0 0 0 )
Sampling this produces (1 1 0 1 1 0 .). Now the average intensity in the
image is still .5 (can you prove that this will always be true) and we have something that
looks more like equal sized white and black stripes.

8. Suppose, for the purposes of this problem, that Philadelphia is 90 miles north-east
of Washington, and that Baltimore is 30 miles northeast of Washington.
a. If it is 40 degrees in Washington, and 31 degrees in Philadelphia, use
linear interpolation to estimate the temperature in Baltimore.

Baltimore is 1/3 of the way from Washington to Philadelphia. So we can express the
position of Baltimore as (2/3)*Washington + (1/3)*Philadelphia. If we express the
temperature in the same way, we get: (2/3)*40 + (1/3)*31 = 37 degrees.

b. Suppose Harrisburg is 60 miles north of Washington, and it is 24 degrees

First, lets consider a point that is 30 miles north of Washington. This is halfway in
between Washington and Harrisburg, and so linear interpolation predicts a temperature of
32 degrees. Next, we can interpolate between Washington and Philadelphia to get the
temperature 30 miles north and 30 miles east of Washington. This is sqrt(18)/9 = .4714
of the way from Washington to Philly. Interpolating, we find the temperature here to be:
35.76 degrees. The point we want is halfway between these two places, so averaging
gives its temperature as: 33.8 degrees.

c. What do these problems have to do with graphics?

This is exactly the way we determine the z value of every point in a triangle, given the
value at the corners.

9. Consider a triangle with corners at (2,2,3) (3,3,4), (3,1,3).

The normal should be orthogonal to the vectors from the first point to the second and
third ones. These vectors are: (1,1,1) and (1,-1,0). Suppose the vector (a,b,c) is
orthogonal to both these. We have a + b + c = 0, and a-b = 0 (which implies a = b). So,
if we pick c = 2, we get a = b = -1, or (-1, -1, 2). We should normalize this by dividing
the value by sqrt(6).

b. What is an equation for the plane that this triangle lies in?

We can say that the inner product between the normal vector and any point on the plane
is a constant, d. So we get:

-x y + 2z = d.
Plugging in (2,2,3), we get: - 2 2 + 6 = d, so d = 2, giving:

-x y + 2z = 2.

We can verify that this equation holds for all vertices of the triangle.

c. What test would the painters algorithm use to decide paint this triangle
before a triangle with vertices at (2,3,2), (3,2,2), (3,3,3.5)?

The min z value of the first triangle is not greater than the max z value of the second, so
this doesnt work. If we take the inner product of the firsts normal with each point on
the new triangle, we get:

<(-1,-1,2), (2,3,2)> = -1 < 2

Note that for the origin:

<(-1,-1,2),(0,0,0)> = 0 < 2.

These are on the same side of the first triangles plane as the origin, so the second triangle
is in front of the first, and we can render it first.