Compute the descriptor size using MSER as interesting keypoints

2.8k views Asked by At

I am working on an image registration method applied to histological images.

I have one question. I would like to use MSER feature detector to detect keypoints on my image. After the MSER contours were retrieved using the MSER function provided by opencv, I calculate the centroid of each contour in order to use it as an interesting point.

If I make a description of the interesting points directly, with a Surf descriptor for example, the size of the descriptor is one, and is not possible to compare them.

Therefore it is necessary to modify the size of the descriptor with a suitable size.

Does anyone have an idea?

Thanks

1

There are 1 answers

0
user2010428 On

Tha answer is very late but I hope it helps someone.

MSER returns you regions and not points. In order to extract descriptors from MSER regions, you have to map the regions to a standard size, say from any elliptical shape to 30x30 pixel circle and then extract descriptor (such as SURF) out of it.

If you use SURF alone then it uses Harris corners as interest points and then places fixed sized windows around it in a scale space.

Matching is done by comparing descriptors usually. You are trying to match keypoints (or interest points) which is different.

Last but not the least, it is unlikely that MSER centroids and SURF interest points can be can occur at the same location since MSER detect homogenous regions and SURF uses Harris Corners. Centroids of MSER can not have corners, so technically speaking, they will always be the outliers for each other.

In MATLAB 2011 onwards, MSER can be combined with SURF descriptors (Included in the Computer Vision System ToolBox).

One other way is to use computer_descriptors.ln shared by: http://www.robots.ox.ac.uk/~vgg/research/affine/descriptors.html

Best Regards