CS194-26 (CS294-26): Image Manipulation and Computational Photography

This project explores gradient-domain processing, a simple technique
with a broad set of applications including blending, tone-mapping, and
non-photorealistic rendering. For the core project, we will focus on
"Poisson blending"; tone-mapping and NPR can be investigated as bells
and whistles.

The primary goal of this assignment is to
seamlessly blend an object or texture from a source image into a
target image. The simplest method would be to just copy and paste the
pixels from one image directly into the other. Unfortunately, this
will create very noticeable seams, even if the backgrounds are
well-matched. How can we get rid of these seams without doing too
much perceptual damage to the source region?

One way to approach this is to use the Laplacian pyramid blending technique we implemented for the last project (and you will compare your new results to the one you got from Laplacian blending). Here we take a different approach. The insight we will use is
that people often care much more about the gradient of an image than
the overall intensity. So we can set up the problem as finding values
for the target pixels that maximally preserve the gradient of the
source region without changing any of the background pixels. Note
that we are making a deliberate decision here to ignore the overall
intensity! So a green hat could turn red, but it will still look like
a hat.

We can formulate our objective as a least squares
problem. Given the pixel intensities of the source image "s" and of
the target image "t", we want to solve for new intensity values "v"
within the source region "S":

Here, each "i" is a pixel in the source region "S",
and each "j" is a 4-neighbor of "i". Each summation guides the
gradient values to match those of the source region. In the first
summation, the gradient is over two variable pixels; in the second,
one pixel is variable and one is in the fixed target region.

The method presented above is called "Poisson blending". Check out the Perez et al. 2003 paper to see sample results, or to wallow in extraneous math. This is just one example of a more general set of gradient-domain processing techniques. The general idea is to create an image by solving for specified pixel intensities and gradients.

So, how do we blend in the swimmers? We construct a new image "v" whose gradients inside the region "S" are similar to the gradients of the cutout we're trying to paste in (swimmers + a little bit of background). The gradients won't end up matching exactly: the least squares solver will take any hard edges of the cutout at the boundary and smooth them by spreading the error over the gradients inside "S".

Outside "S", "v" will match the pool. We won't even bother computing the gradients of the pool outside S; we'll just copy those pixels directly.

In the first half of the large equation above, we set the gradients of "v" inside S. We loop over all the pixels inside the region S, and request that our new image "v" have the same gradients as the swimmers. The summation is over every pixel i in S; j is the 4 neighbors of i (left, right, up, and down, giving us both the x and y gradients. You may notice this equation counts all the gradients twice - is was simpler to write it this way. You actually don't have to double-count everything; it doesn't affect the result.)

But what do we do around the boundary of "S"? I.e. what if i is inside S, but j is outside? That's the second half of the equation. (If you look closely at the variables under the summation sign, the S has a bar thing in front of it - that means the complement of S, or anything not in S. Thus, read "j in the 4-neighborhood of i, but j not in S"). In this case we aren't solving for a v_j, since j is not inside "S". So we just pluck the intensity value right out of the target image. Thus we use t_j. Remember, we aren't modifying t outside the area of S, so we know that v_j in this case isn't a variable: it's exactly equal to t_j.

The implementation for gradient domain processing is not complicated, but it is easy to make a mistake, so let's start with a toy example. In this example we'll compute the x and y gradients from an image s, then use all the gradients, plus one pixel intensity, to reconstruct an image v.

Denote the intensity of the source image at (x, y) as s(x,y) and the values of the image to solve for as v(x,y). For each pixel, then, we have two objectives:

1. | minimize ( v(x+1,y)-v(x,y) - (s(x+1,y)-s(x,y)) )^2 | the x-gradients of v should closely match the x-gradients of s |

2. | minimize ( v(x,y+1)-v(x,y) - (s(x,y+1)-s(x,y)) )^2 | the y-gradients of v should closely match the y-gradients of s |

3. | minimize (v(1,1)-s(1,1))^2 | The top left corners of the two images should be the same color |

The first step is to write the objective function as a set of least squares constraints in the standard matrix form: (Av-b)^2. Here, "A" is a sparse matrix, "v" are the variables to be solved, and "b" is a known vector. It is helpful to keep a matrix "im2var" that maps each pixel to a variable number, such as:

im2var = zeros(imh, imw);

im2var(1:imh*imw) = 1:imh*imw;

(If you find that matlab trickery confusing, understand that you could have performed the mapping between pixel and variable number manually each time: e.g. the pixel at s(r,c), uses the variable number (c-1)*imh+r. However, this trick will come in handy for Poisson blending, where the mapping is from an arbitrarily-shaped block of pixels and won't be such a simple function. So it makes sense to understand it now.)

Then, you can write Objective 1 above as:

A(e, im2var(y,x+1))=1;

A(e, im2var(y,x))=-1;

b(e) = s(y,x+1)-s(y,x);

Here, "e" is used as an equation counter. Note that the y-coordinate is the first index in Matlab convention. Objective 2 is similar; add all the y-gradient constraints as more rows to the same matrices A and b. As another example, Objective 3 above can be written as:

A(e, im2var(1,1))=1;

b(e)=s(1,1);

To solve for

Step 1: Select
source and target regions. Select the boundaries of a region in the
source image and specify a location in the target image where it
should be blended. Then, transform (e.g., translate) the source image
so that indices of pixels in the source and target regions correspond.
I've provided starter code
(getMask.m, alignSource.m) to help with this. You may want to augment
the code to allow rotation or resizing into the target region. You
can be a bit sloppy about selecting the source region -- just make
sure that the entire object is contained. Ideally, the background of
the object in the source region and the surrounding area of the target
region will be of similar color.

Step 2: Solve the blending
constraints:

Step 3: Copy the solved values v_i into your target image. For RGB
images, process each channel separately. Show at least three results
of Poisson blending. Explain any failure cases (e.g., weird colors,
blurred boundaries, etc.).

**Tips**

1. Pre-initialize your sparse matrix with
`
sparse([], [], [], M, N, nzmax)`
for a matrix with M equations and N variables and at most nzmax non-zero entries.

2. For your first blending example, try something that you know should work, such as the included penguins on top of the snow in the hiking image.

3. Object region selection can be done very crudely, with lots of room around the object.

Follow the same steps as Poisson blending, but use the gradient in source or target with the larger magnitude as the guide, rather than the source gradient:

Here "d_ij" is the value of the gradient from the source or the target image with larger magnitude, i.e. `if abs(s_i-s_j) > abs(t_i-t_j), then d_ij = s_i-s_j; else d_ij = t_i-t_j.` Show at least one result of blending using mixed gradients. One possibility is to blend a picture of writing on a plain background onto another image.

** Color2Gray (20 pts) **

Sometimes, in converting a color image to grayscale (e.g., when printing to a laser printer), we lose the important contrast information, making the image difficult to understand. For example, compare the color version of the image on right with its grayscale version produced by rgb2gray.

Can you do better than rgb2gray? Gradient-domain processing provides one avenue: create a gray image that has similar intensity to the rgb2gray output but has similar contrast to the original RGB image. This is an example of a tone-mapping problem, conceptually similar to that of converting HDR images to RGB displays. To get credit for this, show the grayscale image that you produce (the numbers should be easily readable).

Hint: Try converting the image to HSV space and looking at the gradients in each channel. Then, approach it as a mixed gradients problem where you also want to preserve the grayscale intensity.

** More gradient domain processing (up to 20 pts) **

Many other applications are possible, including non-photorealistic rendering, edge enhancement, and texture or color transfer. See Perez et al. 2003 and GradientShop for further ideas.

- Images, including the toy image, sample images for blending, and the color2gray image.
- Starter Code, including a top-level script and functions to select the source region and align source and target images for blending.

Use both words and images to show us what you've done.

Submit all code to bCourses. Include a README describing the contents of each file.

In the website you upload using the upload utility, please:

- Include a brief description of the project.
- Show your favorite blending result. Include: 1) the source and target image; 2) the blended image with the source pixels directly copied into the target region; 3) the final blend result. Briefly explain how it works, along with anything special that you did or tried. This should be with your own images, not the included samples.
- Next, show at least two more results for Poisson blending, including one that doesn't work so well (failure example). Explain any difficulties and possible reasons for bad results.
- Show at least one result for blending with mixed gradients. Again, show the source image, target image, and final result.
- Choose one pair of images that you belnded together using Laplacian pyramid blending in the previous project and now blend it using the Poisson and mixed gradient blending techniques. Display the original images as well as the three blending results. Which approach works best for these images? Why? When do you think that one approach would be more appropriate than another?
- Tell us about the most important thing you learned from this project!

This project was based on one offered by Derek Hoiem.

Here is another description of it from James Hays.