extractFeatures, reference to "Block"-method

Hi

I'm compairing the SURF- and the Block-method in the extractFeatures-function.

Is there any article which describes the Block-method which the Matlab implementation is built upon? Is it the same implementation as the "Block-matching method" in OpenCV? I can't find much (if any) information in the book "Learning OpenCV" which is referred to in the documentation.

Answer 1

[Alex Taylor]

I'm not sure about where to find a good reference at the moment, if you're looking for a paper, but if you look in the reference page for extractFeatures:

http://www.mathworks.com/help/toolbox/vision/ref/extractfeatures.html

The description of block is:

"Simple square neighbhorhood. Note: The Block method extracts only the neighborhoods fully contained within the image boundary. Therefore, the output, VALID_POINTS may contain fewer points than the input POINTS."

In the example that is provided:

    I = imread('pout.tif');

      hcornerdet = vision.CornerDetector;
      points = step(hcornerdet, I);

     [features, valid_points] = extractFeatures(I, points);

They are using the default 'auto' method, which will use the 'block' method because the input points are Nx2 coordinates from the corner detector.

Block feature extraction just extracts small neighborhood blocks of the actual image data centered at the [X Y] locations specified in the point input. In this case, the extracted features blocks of the input image 'pout.tif' centered at corner locations.

Hope that helps.

[Tobias Pahlberg]

Do you know if it's the same method as the "Block-matching method" in OpenCV?

Are the Block-feature points made rotationally invariant like for example SURF or SIFT?