James Besancon (your cs id)

CS 143 / Project 1 / Image Filtering and Hybrid Images

This image filtering algorithm was implemented by taking taking element-wise array products at each pixel of an image after that image had been padded with reflected content on each edge. Examples of of gaussian blur and sobel filters are shown below.

Original

Gaussian Blur Filter

Horizontal Sobel Filter

Hybrid Images were constructed using two images, one of which provides low frequencies (via blur filtering), while the other image provides high frequencies (by subtracting off low frequencies). The sum of the two images makes an image that appears more like the high frequency image at full resolution, but more like the low frequency image as resolution is decreased. Below is a recreation of the given SIGGRAPH example, computed with a filter cutoff frequency parameter of 7 pixels.

High Frequencies

Low Frequencies

Hybrid images work best with highly aligned (in terms of both color and general shape) images, but some fudging can be done. Below is an image showing the SciLi's true form, computed with a filter cutoff frequency parameter of 9 pixels. Though the two images are rather disparate in terms of color, aligning the darker low frequency image with the darker parts of the original high frequency image provides a plausible image.

High Frequencies

Low Frequencies

Images that do not align spatially are trickier to construct. Here you can see two images whose lines do not align, making for a relatively poor hybrid effect in the largest image. However, there is a small succesess in that the last images appear as though David has turned his head.

High Frequencies

Low Frequencies

That being said, some images that are not highly aligned turn out better than you might expect. Here we see a rather pleasing animorphs effect, though the coloration of the boys face is unideal.

High Frequencies

Low Frequencies