What is the maximum filter response in the MR8 filter bank?

So, I am introducing the Maximum Response Filter (MR) Bank - MR8. I use this as a link: http://www.robots.ox.ac.uk/~vgg/research/texclass/filters.html

And so far I have written 38 filters, but I have no idea how to calculate the maximum filter response from 6 turns of each change in the edge and filter strip.

  • What is a filter response? Is this the result of applying a filter or something else?

  • What is the maximum answer?

  • How to calculate maximum responses for edge and bar filters?

+6
source share
2 answers
  • Suppose you have a filter f and image I Filtering I with f is a direct convolution of I and f . Let's say the filtered image is I_f . The response of the filter f to the pixel (x,y) of I is I_f (x,y) .

  • Now you have a series of oriented filters f1 , f2 , f3 , ... Therefore, you will have filtered images I_f1 , I_f2 , I_f3 , ... From them you create the maximum image of the response I_f_max as:

    I_f_max(x,y) = max ( I_f1(x,y) , I_f2(x,y) , I_f3(x,y) , ...)

    Therefore, from the application of 6 oriented filters, you created one response image, which takes into account only the maximum filter response at each pixel and discards the rest.

  • 1 and 2 should answer this.

+5
source

To answer your next question about color images, the VZ classifier specified at https://www.robots.ox.ac.uk/~vgg/research/texclass/with.html uses grayscale images.

However, there are variations that use multi-channel / color images, mostly made by Burghouts and Geusebroek, http://www.bmva.org/bmvc/2006/papers/362.pdf .

However, this adds additional complexity and other options, such as the inclusion of supporting vector machines (SVM) (“On the Importance of Real Conditions for the Classification of Materials”, E. Hayman, etc.) can lead to a larger increase in classification rates, and also provides a level of scale invariance, one of the main limitations in the 2001 VZ classifier.

0
source

All Articles