This project will highlight topics in hybrid blending, multiresolution blending, and poisson blending. Further details can be found at cs194-26/project3.
As a warm up, I thought that I would take my "professional" profile picture and try to make myself look sharper! The process starts by taking the original image and smoothening it via a gaussian filter. The detailed image is than extracted from original - smoothed. Finally we re-add the detailed image (times some alpha "a") back to the original.
Decent photo but it isn't sharp enough to attract LinkedIn recruiters.
We set a = 10 and get a slightly vibrant red on the jacket as a result.
We go with a = 20 and notice that the edges around the face are more prominent.
We push things with a = 100 and realize that too much of a good thing is not good. The image becomes apparently grain-like as seen on the face.
Hybrid images are static images that change in interpretation as a function of the viewing distance. The basic idea is that high frequency tends to dominate perception when it is available, but, at a distance, only the low frequency (smooth) part of the signal can be seen. By blending the high frequency portion of one image with the low-frequency portion of another, you get a hybrid image that leads to different interpretations at different distances.
Back again with this profile picture but this time we are looking to blend me with something else to make me more appealing.
Good ol' log magnitude of Fourier Transform for the first image.
This is the log magnitude of Fourier Transform for the first image after applying the low pass filter.
I figured that since people like dogs, I should blend myself with an Akita.
Good ol' log magnitude of Fourier Transform for the second image.
This is the log magnitude of Fourier Transform for the second image after applying the high pass filter.
This is what you get when you put me and dog together. If you look closely, you might just see the pupper.
This is the log magnitude of Fourier Transform for the hybrid image.
Here's an example of blending Mark Wahlberg and Matt Damon together. They already look the same in real life, so the blend was easy.
Here's an example of blending Joseph Gordon-Levitt and myself. I was hoping to get good results but I guess we only look alike in real life.
For this part, we take an image and create two stacks through continous application of the gaussian filter on the image and then subtracting the original from the filtered image to get the laplacian. Through this process we are able to see the high pass image and low pass image separately.
This is the original image of one of Dali's paintings.
Let's try to deconstruct this failed experiment.
The goal of this part of the project is to blend two images seamlessly using a multi resolution blending. An image spline is a smooth seam joining two image together by gently distorting them. Multiresolution blending computes a gentle seam between the two images seperately at each band of image frequencies, resulting in a much smoother seam.
We start things off with a basic example using an apple as the left side of the blend.
With the orange on the right side, we hope to have a nice blend between the two.
They do look blended though the image turned out darker as a result of not normalizing the pixels.
One can ony dream of the day that these two artists release a joint album. In the meantime, here is a blend of them.
Here is another dynamic duo in hip hop! Let's take a deeper look at the blending.
This project explores gradient-domain processing, a simple technique with a broad set of applications including blending, tone-mapping, and non-photorealistic rendering. For this warmup, 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.
Here we have a grayscale image from the classic movie, Toy Story.
Here the result looks the same as the original.
For this part, we want to seamlessly blend an object or texture from a source image into a target image. The simplest method would be to just directly apply the pixels from one image into the other. Unfortunately, this will create very noticeable seams, even if the backgrounds are well-matched. Instead we will look at a method known as poisson blending which retrieves the gradient of the source image and then blends the intensity values of the outer region from the target image to construct a better blend.
Let's take this touching scene and spice it up.
This scene seems appropriate.
Though it may fool some, this image could potentially be improved. The color of sand around the sand is quite different than that in the movie scene.
An improvement from the previous approach. There are some parts of the image that aren't fully blended. This could be due to some minor changes in color around the border of the source image.
Let's take this cute penguin and place him somewhere.
A snowy region.
The result of the blend went well as expected since the source and target did not differ that much in pixel intensity.
Here is a picture that we will try to paste and blend.
Looks like Harry could use a fellow bro so let's try to add one.
The naive approach has displays very different color values making it not seamless.
Despite the lower half of the image blending in properly,the overall gradient of the source causes the top half to be too dark.
Let's try to do some comparisons with the different blend techniques.
For each blending technique, we will use similar masks such that the results will have some consistency.
In this blend, there is apparant contrast differences and the edges don't seem to blend well at all.
While the contrast was adjusted via the gradient calculations, the edges still look obvious around the face.
While this blend has the largest difference from the original images, it also has better blending both in intensity and seamless edges.