how to train image database save and retrieve results faster?

asked 2014-11-27 00:42:30 -0500

noobproof gravatar image

I have just started with OpenCV. My motivation is object detection. I am using Bag of words algorithm

I am using caltech database. Partially Annotated Databases The Caltech Database

I have attached my working code. So far the recognition is based on MAT() which is formed after looping over all the positive dataset images. So everytime I introduce a new test image out of the training dataset the code build the new MAT(), bowTrainer.add(features), and new Vocabulary.

I want to remove this step of training again and again. So that when i give a new image it checks against "final trained matrix" and gives result quickly.

Also, how to implement it such a way that after predicting the class of input image system is trained afterwards??

#include "stdafx.h"
#include "opencv\cv.h" 
#include "opencv\highgui.h"
#include "opencv\ml.h"
#include <stdio.h>
#include <iostream>
#include <opencv2/features2d/features2d.hpp>
#include <opencv2/imgproc/imgproc.hpp>
#include <opencv2/highgui/highgui.hpp>
#include <opencv2\nonfree\nonfree.hpp>
#include <vector>

using namespace cv; 
using namespace std;

using std::cout;
using std::cerr;
using std::endl;
using std::vector;

char ch[30];

//--------Using SURF as feature extractor and FlannBased for assigning a new point to the nearest one in the dictionary
Ptr<DescriptorMatcher> matcher = DescriptorMatcher::create("FlannBased");
Ptr<DescriptorExtractor> extractor = new SurfDescriptorExtractor();
SurfFeatureDetector detector(500);

//---dictionary size=number of cluster's centroids
int dictionarySize = 1500;

TermCriteria tc(CV_TERMCRIT_ITER, 10, 0.001);
int retries = 1;
int flags = KMEANS_PP_CENTERS;
BOWKMeansTrainer bowTrainer(dictionarySize, tc, retries, flags);
BOWImgDescriptorExtractor bowDE(extractor, matcher);

// this function is being used to train the BOW classifier 
void collectclasscentroids() {
    IplImage *img;
    int i,j;
            sprintf( ch,"%s%d%s%d%s","train/",j," (",i,").jpg");
            const char* imageName = ch;
            img = cvLoadImage(imageName,0);
            vector<KeyPoint> keypoint;
            detector.detect(img, keypoint);
            Mat features;
            extractor->compute(img, keypoint, features);

int _tmain(int argc, _TCHAR* argv[]){
    int i,j;
    IplImage *img2;
    cout<<"Vector quantization..."<<endl;


    vector<Mat> descriptors = bowTrainer.getDescriptors();
    int count=0;
    for(vector<Mat>::iterator iter=descriptors.begin();iter!=descriptors.end();iter++){
    cout<<"Clustering "<<count<<" features"<<endl;

    //choosing cluster's centroids as dictionary's words
    Mat dictionary = bowTrainer.cluster();

    cout<<"extracting histograms in the form of BOW for each image "<<endl;
    Mat labels(0, 1, CV_32FC1);
    Mat trainingData(0, dictionarySize, CV_32FC1);
    int k=0;
    vector<KeyPoint> keypoint1;
    Mat bowDescriptor1;

    //extracting histogram in the form of bow for each image 
            sprintf( ch,"%s%d%s%d%s","train/",j," (",i,").jpg");
            const char* imageName = ch;
            img2 = cvLoadImage(imageName,0);
            detector.detect(img2, keypoint1);
            bowDE.compute(img2, keypoint1, bowDescriptor1);
            //setting the label for images
            labels.push_back((float) j);

    //Setting up SVM parameters
    CvSVMParams params;
    params ...
edit retag flag offensive close merge delete


Here is a code similar to what you want to do. The idea is that you shall have 2 Mat, one of labels and one of predicted labels. And then compare them with a countNonZero.

thdrksdfthmn gravatar imagethdrksdfthmn ( 2014-11-28 02:42:58 -0500 )edit