Map contour points to original image [closed]

asked 2017-07-10 16:27:20 -0500

updated 2017-07-11 08:17:24 -0500

I've performed some basic image operations and managed to isolate the object I am interested in.

As a result, I ended up with a container; get<0>(results) holding the corresponding (x, y) of the object. For visual purposes, I drew these points on a different frame named contourFrame

My question is, how do I map these points back to the original image?

I have looked into the combination of findHomography() and warpPerspective but in order to find the homography matrix, I need to provide respective destination points which is what I am looking for in the first place.

remap() seems to do what I am looking for but I think its a misunderstanding of the API on my part but after trying it out, it returns a blank white frame.

Is remap() even the way to go about this? If yes, how? If no, what other method can I use to map my contour points to another image?

A Python solution/suggestion works with me as well.

Edit I

For the current scenario, no mapping is needed back to the original image. But what if I rotate the original image? Or move my camera back and forth i.e. move it closer and further away? How can I still map back the coordinates to the newly oriented original image?

Original Image

iPad

Results

Results

Top Left: Canny Edge Top Right: Dilated Bottom Left: Eroded Bottom Right: Frame with desired object/contour

tuple<vector<vector<Point>>, Mat, Mat> removeStupidIcons(Mat& edges)
{
    Mat dilated, eroded;
    vector<vector<Point>> contours, filteredContours;

    dilate(edges, dilated, Mat(), Point(-1, -1), 5);
    erode(dilated, eroded, Mat(), Point(-1, -1), 5);

    findContours(eroded, contours, CV_RETR_LIST, CV_CHAIN_APPROX_SIMPLE);

    for(vector<Point>contour: contours)
        if(contourArea(contour) < 200)
            filteredContours.push_back(contour);

    return make_tuple(filteredContours, dilated, eroded);
}

Mat mapPoints(Mat& origin, Mat& destination)
{
    Mat remapWindow, mapX, mapY;

    mapX.create(origin.size(), CV_32FC1);
    mapY.create(origin.size(), CV_32FC1);

    remap(origin, destination, mapX, mapY, CV_INTER_LINEAR);

    return destination;
}

int main(int argc, const char * argv[])
{
    string image_path = "ipad.jpg";

    original = imread(image_path);

    blur(original, originalBlur, Size(15, 15));
    cvtColor(originalBlur, originalGrey, CV_RGB2GRAY);

    Canny(originalGrey, cannyEdges, 50, 130, 3);
    cannyEdges.convertTo(cannyFrame, CV_8U);

    tuple<vector<vector<Point>>, Mat, Mat> results = removeStupidIcons(cannyEdges);
//  
//  get<0>(results) -> contours
//  get<1>(results) -> matrix after dilation
//  get<2>(results) -> matrix after erosion

    Mat contourFrame = Mat::zeros(original.size(), CV_8UC3);

    Scalar colour = Scalar(rand()&255, rand()&255, rand()&255);

    drawContours(contourFrame, get<0>(results), -1, colour, 3);

    Mat contourCopy, originalCopy;

    original.copyTo(originalCopy); contourFrame.copyTo(contourCopy);

    // a white background is returned.
    Mat mappedFrame = mapPoints(originalCopy, contourCopy);

    imshow("Canny", cannyFrame);
    imshow("Original", original);
    imshow("Contour", contourFrame);
    imshow("Eroded", get<2>(results));
    imshow("Dilated", get<1>(results));
    imshow("Original Grey", originalGrey);
    imshow("Mapping Result", contourCopy);

    waitKey(0);

    return 0;
}
edit retag flag offensive reopen merge delete

Closed for the following reason the question is answered, right answer was accepted by sturkmen
close date 2017-09-02 16:58:57.700493

Comments

why do you think, this needs any mapping at all ? the contours are coords in the original image. there is no transformation of any kind involved.

(also, remapping would require some content in mapX and mapY)

berak gravatar imageberak ( 2017-07-11 04:55:14 -0500 )edit

@berak It was a question clarification on my part. Check the edit

eshirima gravatar imageeshirima ( 2017-07-11 08:19:39 -0500 )edit

apologies, i misread it all. you're takiing a photo of your tablet, right ? (so ofc. there's some camera angle.)

i mistook it for a screenshot.

berak gravatar imageberak ( 2017-07-11 10:19:12 -0500 )edit

Its fine no worries.. Any recommendations on how to go about mapping the points?

eshirima gravatar imageeshirima ( 2017-07-11 10:32:33 -0500 )edit

this is basically the same job as finding corners of a paper or a form.

(get corners, getPerspectiveTransform(), warpPerspective())

just look up "sudoku solver opencv" or similar, and you'll fnd tons of examples.

berak gravatar imageberak ( 2017-07-11 10:37:18 -0500 )edit