Examples

First steps - Tutorial

1. Take a a gray level image of yours (e.g. a.pgm)
2. Interest Point Detection (with Harris-Laplace):

   Detect -dt har a.pgm a.har

   Note: if you use -affine detection also specify -angles

3. Compute local descriptors for each interest point (based on sift)

   ComputeDescriptor a.pgm a.har a.har.sift

4. Show result of (2) on the image:

   DrawCorners -col y -wi 2 a.pgm a.har a.har.pnm

5. Dump the result of (3) to a text

   dumpcontents a.har.sift > a.txt

6. Dump only the computed descriptors (of step (3)) to a text

   corners2text a.har.sift -adddesc a.har.desc.txt

7. Convert the results to MatLab format:

   corners2matlab a.har.sift a.mat

Detection with a list of images

1. Create a file called images.lst. Each line should contain a name of an image.
2. Detect interest points based on Laplace-of-Gaussian.

   Detect -dtype log @images.lst result.log

3. Apply affine and rotation invariance for each point and compute SIFT descriptors. (the next two lines should be typed in one!)

   ComputeDescriptor -dt sift -aff -an @images.lst result.log   result.log.sift

4. Draw the detection results (with affine estimation) on the 3^rd image in the list.

   SelectCorners -fid 2 result.log.sift third.sift   DrawCorners -color yellow @images.lst third.sift third.pnm

   Note that first we have to select the points of the third image to a different file (fid==2, because the files are numbered from 0), then we can use DrawCorners with the list file, it will load up the correct image.

5. Show the same patches but now extracted in a thumbnail format.

   DrawCorners -th_on -co ye @images.lst third.sift third_th.pnm

Examples for SelectCorners

Select 10 random interest point:

Select interest points having scale between 2 and 10:

Select the 10^th interest point:

Select the 10^th and 20^th interest points: