CS194-26 Proj 2

Kehan Wang

Part 1

1.1

The magnitude of gradient is computed by:
$$|g| = \sqrt{g_{x}^{2} + g_{y}^2}$$
where g_x, g_y comes from convolving D_x, D_y against the image.

We then see every pixel of the convolution results as a gradient in x or y direction - g_x or g_y, and compute the magnitude using the above formula.

1.2

The difference we see after applying guassian kernel is that the noise in the background is removed. In 1.1, we can see small white dots in the lower part of the background. In 1.2, there is less noise.

1.3

Official image

• Rotated

• Rotation Corrected => -3 degree

• Histograms

Self-select image1: success

• Original

• Add rotation -10 degrees

• Rotation Corrected => 10 degree

• Histograms

Self-select image2: success

• Original

• Add rotation 17 degrees

• Rotation Corrected => -17 degree

• Histograms

Self-select image3: failure

• Original

• Add rotation -14 degrees

• Rotation Corrected => -6 degree

• Histograms

2.1

Official image

• Original

• Blurred

• Sharpened

Self-select image

• Original

• Blurred

• Sharpened

2.2

Official image

• Original

  • Derek

    Image	Fourier

  • Cat

    Image	Fourier

• Filtered

  • Derek Low Frequency

    Image	Fourier

  • Cat High Freqency

    Image	Fourier

• Hybrid

  Image	Fourier

Self select image 1 - success

Trump	Kim Jong-Un	Supreme Leader

Self select image 2 - success

Apple	Laugh	Applaugh

Self select image 3 - failure

SpaceX rocket is not good enough to be a superman yet.

Superman	Rocket	Rocketman

Bell and whistles:

I used grayscale for low frequency and color for high frequency so the color for low frequency does not dominate the view.

2.3

• Hybrid image

• Self select image

2.4

• Oraple

  Orange	Apple	Oraple

• Itachi Uchiha (irregular mask)

  Sharingan	Eye	Itachi

• Lionman

  Man	Lion	Lionman