IMAGE Team - The Pantheon project

---

Tutorial: Detecting Template in Image

Identification

• Pantheon Project
• Author: Régis CLOUARD
• Date: May 29, 2013
• Required: Pandore
• Optional: Ariane

Resources

• The images
• The Ariane workspace
• Notice: This script needs two additional operators (to be compiled with Pandore).
  • pcrosscorrelation.cpp
  • pgetwindowaroundpoints.cpp

  I. Objective

  The objective is to determine the position of all instances of a given template in an image. The template is given as a separate image.

  The source image.	The template.	The detected instances of the template.

  II. Method

  The method is based on the cross correlation. Cross-correlating an image with a known template gives high values wherever the template is a match.

  The method is quite simple :

  1. Apply the cross-correlation of the input image by the template.
  2. Localize the maxima of the correlation.
  3. Construct the result of the detection.

  II.1. Cross-correlation of the Image by the Template

  The cross correlation measures for each pixel the degree of matching between the template and the image centered at the pixel. The correlation value is between [0..1], where 1 corresponds to a perfect match.

  The correlation is implemented by convolution:

  

  where t is the mean value of template image and f_u,v is the mean value of the image pixels inside the window around the pixel (u,v).

  pany2pan page.png page.pan pcrosscorrelation page.pan letter.pan image1.pan
  	image1.pan : The result of the cross-correlation seen as level map.

  II.2. Localization of the Maxima of the Correlation

  The localization consists in finding all the pixels that are local maxima and have a correlation value greater than 0.7 out of 1.

  pabs image1.pan tmp1.pan plocalmaxima 8 tmp1.pan tmp2.pan pbinarization 0.7 1 tmp2.pan image2.pan
  	image2.pan : The peaks of cross-correlation.

  II.3. Construction of the Result

  For the sake of visualization, the output image is built with the pixels of the input image inside a window around the maxima. The size of the window is the size of the template image.

  i) Unix/Linux (bash) pproperty 0 letter.pan; ncol=`pstatus` pproperty 1 letter.pan; nrow=`pstatus` pgetwindowaroundpoints $ncol $nrow 0 page.pan image2.pan image3.pan ii) Windows (MsDos) pproperty 0 letter.pan call pstatus call pset ncol pproperty 1 letter.pan call pstatus call pset nrow pgetwindowaroundpoints %ncol% %nrow% 0 page.pan image2.pan image3.pan ppan2png image3.pan output.png
  	image3.pan : The detected instances of the template.

  III. The Complete Pandore Script (bash)

  Code

  pany2pan page.png page.pan
  pcrosscorrelation page.pan letter.pan image1.pan
  
  pabs 8 image1.pan tmp1.pan
  plocalmaxima 8 tmp1.pan tmp2.pan
  pbinarization 0.7 1 tmp2.pan image2.pan
  
  pproperty 0 letter.pan; ncol=`pstatus`
  pproperty 1 letter.pan; nrow=`pstatus`
  pgetwindowaroundpoints $ncol $nrow 0 page.pan image2.pan image3.pan
  ppan2png image3.pan output.png 
  

  ---

  The Pantheon project
  Image Team GREYC Laboratory
  UMR CNRS 6072 - ENSICAEN - University of Caen, France
  This page was last modified on 14 November 2013