The goal of this project is to create hybrid images using the approach described by the SIGGRAPH 2006 paper by Oliva, Torralba, and Schyns.

The algorithm takes in 2 images, aligns them, and passes 1 image through a high-pass filter and another through a low-pass filter. For the low-pass filter, we used a standard 2D Gaussian filter. For the high-pass filter, we simply subtracted the Gaussian filtered image from the original.

Shanghai/NYC

Original Images

Source https://www.flickr.com/photos/freshwater2006/14955547241/in/photolist-oMz69M-os1nwy-os16LX-oEHqnu-oQNyAK-p5fb3R-p5vgCx-p8jzKi-89bedR-6puKfD-cor64L-6pz7fA-oEHNuB-oovBMz-oE96De-odhbAg-pcKxu8-ormAHh-p5Qsem-oMv7cb-cdDKJL-5vi7Vc-bjRore-jsPxZm-fdJgDo-5UaDCM-8KLHGq-74EpzM-bweBJg-6pzaRE-6puStc-6p1KWZ-dzJsqd-nVTSvw-85DHWz-6U9NkW-6JZdN6-Dez7i-5Fm7J5-9WCj6U-8hndWe-4N9WhM-tEDR9-aJUJ96-dzJsN7-7tr4W5-4VfXk4-5MLHZn-dzCZ7g-dzCYEx
License

Source: https://www.flickr.com/photos/quintanomedia/15340488382/in/photolist-oVf1E1-p9zBaC-opd1hb-pnA1K1-oAgye4-oLQfNn-ouC3Pi-oZbNRn-pheqHr-o7UzZN-oTR9jn-po6p9e-opYvvS-ofMWC5-oNnU4s-obRmTZ-o893ya-oeGykW-p4k1cq-ougnqU-o6JJrT-pfufwc-oYfVRg-p67pJX-p17RFB-ooz9a9-oSYJCp-p58KZw-pmBaFm-pdwzEv-oWgR12-oWhdzB-p8qVPW-oEN1Jk-pdJtvy-oWix9u-pjAD9q-oZwASx-ofsNmH-o9LiCF-o9Libi-o9KgpW-oferA5-oqXzPV-owm1kP-oGGwGd-oreiBo-oGAaFE-owmq6z-ougA8Y
License

FFT Images

Shanghai

NYC

As we can see from the FFT images, the image of Shanghai contains more high frequency information. Therefore, we passed it through the high-pass filter and passed the NYC image through the low-pass filter. For the filters, we used standard 2D Gaussian filter with sigma = 5.

Filtered Input Images

Shanghai

NYC

FFT Images of Filtered Input Images

Shanghai

NYC

Hybrid Image

Hybrid FFT Image

The result is very good, thanks to the vast difference in frequency information of the 2 images.

Superman Plane

"It's a Bird... It's a Plane... It's Superman!"

Original Images

Source: http://3.bp.blogspot.com/-eaGmFVZ0cNE/T991eBTWvSI/AAAAAAAAAIQ/gA-B18i2nd8/s1600/1673558-superman_fly1.png

Source: https://c1.staticflickr.com/9/8357/8403703984_325c0d0614.jpg

FFT Images

Superman

Plane

Filtered Input Images

Superman

Plane

FFT Images of Filtered Input Images

Superman

Plane

Hybrid Image

Hybrid FFT Image

The result for the superman plane image is good as well since the 2 images have pretty different frequency information.

Old Boat New Boat

Original Images

Source: http://1.bp.blogspot.com/-O1OMXCdY10c/T0-3qjhiGsI/AAAAAAAAAd8/jf2It-DSwiY/s1600/00003.jpg

Source: http://3.bp.blogspot.com/-z7I6t_sdFi0/U3CiW0ViM7I/AAAAAAAAxUM/xgjeu7AMiIA/s1600/Houbei-class-fast-attack-firing-yj-83-anti-ship-missile.jpg

FFT Images

Old Boat

New Boat

As shown by the FFT images, we can see that the image of the new boat contains slightly more high frequency information. Therefore, we are going to pass the new boat image through the high-pass filter and the old boat image through the low-pass filter. Again, we are using a standard Gaussian with sigma = 5.

Filtered Input Images

Old Boat

New Boat

FFT Images of Filtered Input Images

Old Boat

New Boat

Hybrid Image

Hybrid FFT Image

This hybrid image did not work out so well because the 2 images have similar frequency infomation. Therefore, we cannot really distinguish 1 image from the other once their low/high frequency information has been removed. In the case of this image, you can't really since the hull of the new boat in the hybrid image.

I implemented a color version of the hybrid image by retaining the color in the image with more high frequency information (aka the one passed through high-pass filter).

I didn't have time to fine tune the parameters yet, so some of the colors may not be clear in the image.