Image Projections Kavalieratou.

Hi Guys! I have a problem on Structural Characteristics(kavallieratou) technique. Well, i'm implementing this method on matlab for a research, but i'm really new on matlab.(never programed in, never. Only java, but my research needs matlab). I'm have a trouble on step of the projections( horizontal projections, Vertical projections and principally radial projections and profiles(in-out, out -in). In kavallieratou, these steps are necessary to analyze it, and after puts all of them in a vector with 280 characteristics. Going deeper, this technique is to count the black pixels of a row / column, and stores in a vector. The radial projection is calculated by sum of black pixels in the image from the center Geometric image and moving toward the edge pre-defined angles.I just need it for make a comparation with another technics, and i implemented all of others steps but this step i dont know how to do on matlab. Any one knows it?

thanks in advance!

Answer 1

[Image Analyst]

To "count the black pixels of a row / column, and stores in a vector." you'd do this

[rows columns] = size(grayScaleImage);
countOfNonZerosInARow = zeros(1, rows);
countOfNonZerosInAColumn= zeros(1, columns);
for row = 1 : rows
  oneRow = grayScaleImage(row, :);
  countOfNonZerosInARow(row) = nnz(oneRow);
end
for column = 1 : columns
  oneColumn = grayScaleImage(:, column);
  countOfNonZerosInAColumn(column) = nnz(oneColumn);
end

There are other ways, but this might be the most straightforward for a beginner like you. Can you handle it from there?

[Marcus Vinicius]

yes, that was very helpful. With that I can implement the technique of vertical and horizontal projection of a database (images). The technique complete, is to use these two, another technique which is as follows: In an image from eg 16x16 pixels, classifying the geometric center of it and from there, to count the black pixels within an angle of 5 °, the angle from 0 °. This is really breaking my head. The geometric part of the point that the center is to be easy to find. But counting the pixels in that angle, I believe I have to use Fourier coefficients, something like that. I appreciate the help for that horizontal and vertical: D

there goes the article base, for anyone interested :)

http://citeseerx.ist.psu.edu/viewdoc/summary?doi=10.1.1.97.3020

[Marcus Vinicius]

in addition, would be very useful to know how to count pixels that are contained in an angle, you have the coordinates of the center of the image.

[Image Analyst]

You'd probably have to use imrotate to get a rectangular array, then use the code I gave you on that. But you'll have to alter imrotate to make sure it rotates/exposes something other than 0's when it does it's thing.

[Marcus Vinicius]

i will do this way. If i get a good result i post here . Thanks I.A. :D

Answer 2

[Marcus Vinicius]

no one else?

[Image Analyst]

They've probably never heard of the "kavallieratou technique" - I haven't. There are thousands of papers that other people name after the authors - no one can know them all.

[Marcus Vinicius]

I'm sorry, I should have been a little more specific about my problem and should not have named the name of the author of the article that I research.