OpenCV Detection of Dark Objects Against Light Backgrounds

The results of some experimentation with and comparison between raw OpenCV functions and the cvBlobsLib library to detect darker coloured spots against lighter backgrounds.

For guidance on how to set up OpenCV / cvBlobsLIb for use in Visual Studio Environments, take a look at this post or this one.

Main program loop:

// DetectBlobs.cpp : Defines the entry point for the console application.
//
#include "stdafx.h"
#include "BlobResult.h"

#include <cv.h>
#include <cxcore.h>
#include <highgui.h>

int _tmain(int argc, _TCHAR* argv[])
{
    // Initialise
    //std::string filepath = "spots.bmp";
    std::string filepath = "spots2.bmp";
    CBlobResult blobs;
    int num_blobs = 0;

    // Load grayscale version of coloured input image
    IplImage* original  = cvLoadImage( filepath.c_str() );
    IplImage* grayscale = cvLoadImage( filepath.c_str(),
                                       CV_LOAD_IMAGE_GRAYSCALE );

    // Check bitmap image exists
    assert( grayscale );			

    // Create IplImage struct for a black and
    // white (binary) image
    IplImage* img_bw = cvCreateImage( cvGetSize( grayscale ),
                                      IPL_DEPTH_8U,
                                      1 );

    // Use thresholding to convert grayscale image
    // into binary
    cvThreshold( grayscale,             // source image
                 img_bw,                // destination image
                 40,                    // threhold val.
                 255,                   // max. val
                 CV_THRESH_BINARY );    // binary

    // Create IplImage struct for inverted black
    // and white image
    IplImage* img_bw_inv = cvCloneImage( img_bw );
    IplImage* img_bw_cpy = cvCloneImage( img_bw );	

    // Find connected components using cvBlobsLib
    blobs = CBlobResult( img_bw, 0, 255 );

    // Exclude all blobs smaller than the given value
    // The bigger the last parameter, the bigger the
    // blobs need to be for inclusion
    blobs.Filter( blobs,
                  B_EXCLUDE,
                  CBlobGetArea(),
                  B_GREATER,
                  3 );

    num_blobs = blobs.GetNumBlobs(); 

    // Find connected components using OpenCV
    CvSeq* seq;
    CvMemStorage* storage = cvCreateMemStorage( 0 );
    cvClearMemStorage( storage );

    // cvFindContours the 12 + 1 extra object for
    // white backgrounds and black spots, hence
    // subtract 1
    num_blobs = cvFindContours( img_bw,
                                storage,
                                &amp;seq,
                                sizeof( CvContour ),
                                CV_RETR_LIST,
                                CV_CHAIN_APPROX_NONE,
                                cvPoint( 0, 0 ) ) - 1;

    // Display the input / output windows and images
    cvNamedWindow( "original" );
    cvShowImage( "original", original );	

    cvNamedWindow( "grayscale" );
    cvShowImage( "grayscale", grayscale );		

    cvNamedWindow( "black_and_white" );
    cvShowImage( "black_and_white",
                  img_bw_cpy );			

    // Wait for user key press and then tidy up
    cvWaitKey(0);

    cvReleaseImage( &amp;original );
    cvReleaseImage( &amp;grayscale );
    cvReleaseImage( &amp;img_bw );
    cvReleaseImage( &amp;img_bw_inv );
    cvReleaseImage( &amp;img_bw_cpy );

    cvDestroyWindow( "greyscale" );
    cvDestroyWindow( "black_and_white" );
    cvDestroyWindow( "inverted" );

    return 0;
}

Output windows obtained for the original, grayscale and binary images respectively were as follows:

Spot detection using cvBlobsLib

Object detection using the cvBlobsLib is a straightforward matter of creating a CBlobResult object, by giving it the input black and white image and background colour (255); and detecting the blobs themselves by means of the filter method, which in this example is fairly self-explanatory: exclude any blobs whose area is less than 3; output the number of blobs detected:

// Find connected components using cvBlobsLib
blobs = CBlobResult( img_bw, 0, 255 );

// Exclude all blobs smaller than the given value
// The bigger the last parameter, the bigger the blobs need
// to be for inclusion
blobs.Filter( blobs,
              B_EXCLUDE,
              CBlobGetArea(),
              B_GREATER,
              3 );

num_blobs = blobs.GetNumBlobs();

Spot detection using OpenCV

Using raw OpenCV calls to detect and count the objects is achieved with the cvFindContours function as shown in the following code section:

// Find connected components using OpenCV
CvSeq* seq;
CvMemStorage* storage = cvCreateMemStorage( 0 );

cvClearMemStorage( storage );

// cvFindContours the 12 + 1 extra object
// for white backgrounds and black spots, hence
// subtract 1
num_blobs = cvFindContours( img_bw,
			    storage,
			    &seq,
			    sizeof( CvContour ),
			    CV_RETR_LIST,
			    CV_CHAIN_APPROX_NONE,
			    cvPoint( 0, 0 ) ) - 1;

For some reason as yet unclear to me, the cvFindContours returns a count of 13 objects: 12 + one extra, so I have subtracted one in this instance to give the true count. For images with black backgrounds and white objects this seems to be unnecessary.

Downloadable Visual Studio 2008 Project here.

Other posts related to image detection

Tracking Coloured Objects in Video using OpenCV
Displaying AVI Video using OpenCV
Analyzing FlyCapture2 Images obtained from Flea2 Cameras
Integrating the FlyCapture SDK for use with OpenCV
Getting Started with OpenCV in Visual Studio
Object Detection Using the OpenCV / cvBlobsLib Libraries