CS 194-26 Image Manipulation and Computational Photography

Part 1.1 :

The goal of this part is to sharpen images using the unsharp masking technique covered in the class. The idea is to subtract the smoothed image from the original image then add it back to the original image with some scaling factor. Here are some of the result.

Doe Library, original and sharpened

Here is another example:

Boston at night, original and sharpened

You can see there is a clear improvement in the image, as the intensity of the lights on buildings increases. Since different scaling factories and filter kernels give different results, we can experiment with those hyperparameters and get better images. Here is an example of an image and its sharpened versions.

Moscow at night, original and sharpened versions

Part 1.2:

For this part, the goal is to create hybrid images using high and low-frequency fusion. In the end, the result should have two different appearances if viewed from close and far distances. I combined two controversial figures in one image. For instance here is an image of Jesus and Devil(according to google searches) and their combined version. Try looking at the images from different distances.

Jesus and Devil

Hybrid version in gray

You can also see the Fourier transform for each of the filters(high and low) and for the hybrid image in gray

Fourier transforms (Hybrid, low, high)

We can perform the same thing for colored images as well, for each channel individually then combining them together.

Trump and Putin

Hybrid version in color

Part 1.3:

In this part we examine Laplacian and Gaussian Stacks for the previous results. Let's look at the Gaussians first. Here is the stack elements for Jesus -> Devil hybrid image.

The Gaussian Stack

And here is the Laplacian stack for Putin->Trump hybrid. This is a little bit dark as expected.

The Laplacian Stack

Part 1.4:

The goal of this part of the assignment is to blend two images seamlessly using a multi resolution blending. Here are some results.

Apple and Orange

Oraple

Tree in different seasons

Blend of seasons

Part 2.1 and 2.2:

Here we solve a system of linear equations where the variables are the pixels in the blending region. You can see the result of the toy problem below.

Origianl and Resulted images

And here is an example of image blending for the sample image provided.

Source, Target and Mask

Blend