CODEWITHSUNDEEP
opencvimage-processingcomputer-vision
robot learning
robot learning

Reputation: 1

HOG preprocessing normalization

I don't understand what the purpose is of the preprocessing step in HOG (histogram of oriented gradients). In step 1 we normalize the image using a square-root normalization. What is the advantage of this step? Also we have a block normalization. Are these two steps the same?

Upvotes: 0

Views: 797

Answers (1)

globalex
globalex

Reputation: 569

This is easy. Normalization is used to remove local light differences. The pattern in dark scene can be same in light scene but the values are different. [2 2 3 3] is one edge from 2 to 3. [4 4 6 6] is basically same edge 2 times first vector. These two vectors are linearly dependent. Normalization is way to find match these vectors which describes the same in different conditions. First vector l2 norm Sqrt( pow(2)+pow(2)+pow(3)+pow(3)) = 5,09 , the second is sqrt(pow(4)+pow(4)+pow(6)+pow(6)) is = 10,19. If you divide each element of first vector by 5,09 and each element of second vector by 10,19 the result is [0.4 0.4 0.6 0.6]. They are describing the same with different light conditions. This is the basics of Algebra. my blog with cv resources

Upvotes: 2

Related Questions