第5次课，从第6节开始

Recap so far

  1. Convolution
  2. Correlation
  3. separable
  4. Filtering

1 Introduction

Deep learning in frequency domain

Adapted from browncsci1430

  1. Spatial frequency
  2. Fourier transform and frequency domain
    1. Frequency view of filtering
    2. Hybrid images
    3. Sampling
  3. Reminder: Textbook
    1. Today’s lecture covers material in Klette 1, 2; Szeliski 3.4

2 Sampling 采样

2.1 Making lower resolution images

Subsampling by a factor of 2: Throw away every other row and column to create a 1/2 size image

Untitled

im_small = image[::2, ::2 ]

Why does a lower resolution image still make sense to us?  What information do we lose?