1. Martin Ovesný
  2. SIFT

Wiki

Clone wiki

SIFT / Home

Welcome

This page is dedicated to my semestral project for Digital Image Processing class at FIT CTU in Prague.

Info

  • Final report (CZ) PDF
  • Presentation (CZ) PDF

Results

Note: don't ask me why those arrows have different colors :) that's most likely some CImage magic stuff that I can't find out how to turn off


  • Image: smrtka_gsc.jpg
  • Contrast treshold: 0.03
  • Corner treshold: 10.0
  • Description: These are default values that Lowe proposed in his paper.
    • The smaller contrast treshold is, the more key-points will be found.
    • The smaller corner treshold is, the less key-points will be found.
  • smrtka-result.jpg

  • Image: kluk.jpg
  • Contrast treshold: 0.01
  • Corner treshold: 50.0
  • Description: Here I tried to maximize number of the key-points to get some of them on the boy's face, because as you can see, there is not many local extremas on his face. Unlike the background behind his head, which is very "rugged".
  • kluk-result.jpg

  • Image: sift-scene.jpg
  • Contrast treshold: 0.03
  • Corner treshold: 10.0
  • Description: In this image you can see how SIFT key-points are sensitive on high frequencies in two different directions - the trees and the stones on the image are crowded by key-points, unlike the obelisks that are almost clean.
  • sift-scene-result.jpg

  • Image: stadion_gsc.jpg
  • Contrast treshold: 0.08
  • Corner treshold: 5.0
  • Description: In this image I've set the contrast treshold very high to obtain the most significant corners in it. As you can see, key-points are always located in corners where are significant changes of intensities (black <--> white).
  • stadion-result.jpg

  • Image: smile.png
  • Contrast treshold: 0.03
  • Corner treshold: 10.0
  • Description: This should demonstrate scale&rotation invariance.
  • smile_gsc-result.jpg smile_gsc_360-rol-result.jpg


  • Description: This example shows image searching algorithm (1NN+SIFT). All SIFT key-points were generated with default treshholds (0.03,10.0).
  • Database:
    • db.png
  • Example 1:
    • Request:
      • search_car.png
    • Result:
      • result_car.jpg
  • Example 2:
    • Request:
      • search_face.jpg
    • Result:
      • result_face.jpg

Updated