CS143 Project 1 — Jeffrey Pfau (jpfau)

Algorithm

The algorithm required creating pyramids of Laplacian-filtered versions of both images, and, less importantly, Gaussian-filtered versions of both images. This was done simply filtering the image with a small Gaussian filter, storing this version in the Gaussian pyramid, and then subtracting the Gaussian from the original version, yielding the Laplacian version, which is stored in the Laplacian pyramid. The resulting Gaussian is then downsampled by 50%, and the process is then repeated until we have a satisfactory number of elements in the pyramid (in this version, 6).

The second part of the algorithm required reconstructing the hybrid image from the filtered versions. This is accomplished by adding up the Laplacian versions of one image from the highest sampling to the cutoff frequency (generally, the 3rd size), and then taking the rest from the other image's Laplacian pyramid. This, after cropping, yields the images seen.

Results

Provided images

+ =
+ =
+ =

Found images

+ =
+ =
+ =

Discussion

Results are mixed. For some images, this works very well. However, The images have to be selected carefully for this. For many images selected slightly less well, the results are acceptable, but not particularly convincing. However, this might partially be because of flaws with the implementation. The results appear generally more washed out than the examples, leading to a less convincing version of the high filter pass.