Summary Projections

A snapshot of the summary - Projections

• 01 Introduction

• Why visual search engine?

  -   Growing data requires more efficient solution

  
  

  -   Manual indexing is costly and time-consuming

  
  

  -   New technologies are needed to automate processes and to unlock possibilities of big data

  
  

  -   Metadata standards are needed

• 1 Image Filtering

• What is an edge (derivative)?

  An edge is a place of rapid change in the image intensity function
• What is the computational complexity advantage for a separable filter of size k x k, in terms of number of operations per output pixel?

  For a k x k Gaussian filter, 2D convolution requires k^2 operations per pixel

  
  

  But using the separable filters, we reduce this to 2k operations per pixels (3 from top to bottom and 3 from left to right)
• What is the difference between cross-correlation and convolution?

  Flip the filter (a, b, c...) > (i, h, g)
• Why do you use the Gaussian kernel?

  To find edges. 

  
  

  The image is made smooth so the edges are better to be seen.
• How to filter noise?

  1.   Let's replace each pixel with an average of all the values in its neighbourhood.

  
  

  2.   Apply a gaussian filter.
  

  
  

  Correlation filtering ( G = H (X) F)

  Convolution (G = H * F)

  
  
• What is f(x, y)?

  It gives the intensity at position (x, y)
• What is a color image?

  Three functions pasted together 

  
  

  f(x,y) = [r(x,y) g(x,y) b(x,y)
• What is (vector) quantization?

  -   The process of clustering features

  -   Building the visual vocabulary

  
  
• Name three different types of noise

  Salt and pepper noise (white and black pixels)

  
  

  impulse noise (white pixels)

  
  

  Gaussian noise (variations in intensity drawn from a Gaussian normal distribution