Computational Photography Project 2: Michael Kamm (mkamm)

Hybrid Images


For this project, I had to create a hybrid image out of two input images. The closer you look at the combined images, the more it looks like one of the input images and the further away you look, the more it looks like the other input image. This is done by filtering out the high frequencies from one of the images, and filtering out the low frequencies from the other image. After that I blend the two filtered images together.

To make the hybrid images was fairly simple. I used a Gaussian filter to filter out the high frequencies of image 1. I used the same filter to filter out the high frequencies of image 2. To filter out the low frequencies of the 2nd image, I just subtracted the filtered image 2 from the original input image 2. After that I just combined the two filtered images (image 1 with low frequency and image 2 with high frequency).

The only problem was choosing the correct images to make a hybrid image out of and choosing the correct hsize and stddev for the gaussian filter. These were fixed after mindless experimenting and tweaking.

My Favorite Result: Emma Watson and Richard Dawkins

Original Images

When this assignment was announced, I knew exactly which two images I wanted to use. I remember seeing these two images on the internet and noticed the similarity between the way both of their heads are tilted, and their facial expressions. I thought it turned out quite nicely. My Gaussian filter for this was an hsize of 25 and a stddev of 8.33 . I used Emma as my high frequency image and Richard as my low frequency image. I aligned the images by their pupils

Here are the FFT images of Richard Dawkins and Emma Watson, respectively.

Lowpass filter applied to Mr Dawkins

As you can see in the FFT, there are much more low frequencies than high ones in this filtered image.

Highpass filter applied to Ms. Watson

As you can see in the Fourier Domain, there are much more high frequencies than low ones in this filtered image.

The Hybrid image

Here is the FFT of the hybrid image:

Bells and Whistles

I used color to enhance the effect of the hybrid image. I liked the effect of using color on both images the best.

Another Hybrid Image

Another Hybrid Image that I found worked surprising well was the combination of Ainsley Harriott and Bob Ross. My two heroes. For my gaussian filter, my hsize was 27 and my stddev was 9. I aligned the images by their pupils

Original Images


Failure: Sarah Jessica Parker and Sarah Jessica Parker

A hybrid image that did not work too well for me was the combination of two images of Sarah Jessica Parker. One at the red carpet VMAs, and the other at her ranch in the town of Tuscaloozapapolis, Connecticut. I believe the reason that this failed was because in the ranch picture, the image is closer up on her face. Also, there is a huge difference in color between the two images. At the VMAs, Sarah Jessica Parker looks pale, while at her ranch, she has a deep, brown tan. Because the image is closer to her face at the ranch, I aligned the VMA Sarah Jessica Parkers ears, with the Ranch Sarah Jessica Parkers eyes. My hsize was 27 and my stddev was 9.

Original Images