Javacv: Load image from jar.

[mavriksc]

I found this code and made some updates to it.
It allows for easily loading an image from the package to a Mat.
I haven't tested the part where it reverts to use JavaIO to load the bytes.
The issue i think i was getting from loading images was a leading slash and this fixes that.

The constants for color v grey should probably be set with constants from JavaCv but i didn't want to dig thru source at that time. I'm not sure if you want to integrate this or if it's likely to break from updates, or if i may have missed this functionality somewhere else.
Thank you.

//http://answers.opencv.org/question/10236/opencv-java-api-highguiimread-and-paths-pointing-to-files-in-a-jar/
//credit to Lucky Luke 
//i made some updates but i'm not sure about the JavaIO part 
//  as nothing i've loaded used that code yet

import org.bytedeco.javacpp.opencv_core.Mat;
import org.bytedeco.javacpp.indexer.ByteIndexer;

import static org.bytedeco.javacpp.opencv_imgcodecs.*;

import java.awt.image.BufferedImage;
import java.io.IOException;
import java.net.URL;

import javax.imageio.ImageIO;

public final class OpenCVUtils {
    private OpenCVUtils() {}
     /**
    * 8bit, 3-channel image.
    * @see Highgui.CV_LOAD_IMAGE_COLOR
    */
    public static final int LOAD_COLOR = 3;

    /**
    * 8bit, 1-channel image.
    * @see Highgui.CV_LOAD_IMAGE_GRAYSCALE
    */
    public static final int LOAD_GRAYSCALE = 1;

    public static Mat readImage(String name, int flags) {
        URL url = ClassLoader.getSystemResource(name);

        // make sure the file exists
        if (url == null) {
            System.out.println("ResourceNotFound: " + name);
            return new Mat();
        }

        String path = url.getPath();

        // not sure why we (sometimes; while running unpacked from the IDE) end 
        // up with the authority-part of the path (a single slash) as prefix,
        // ...anyways: Highgui.imread can't handle it, so that's why.
        if (path.startsWith("/")) {
            path = path.substring(1);
        }

        Mat image = imread(path,flags);
        if (image.empty()) {
            System.out.println("normal loading failed WHYYY???");
            BufferedImage buf;

            try {
                buf = ImageIO.read(url);
            } catch (IOException e) {
                System.out.println("IOException: " + e.getMessage());
                return image;
            }

            int height = buf.getHeight();
            int width = buf.getWidth();
            int rgb, type, channels;

            switch (flags) {
                case LOAD_GRAYSCALE:
                    type = CV_LOAD_IMAGE_GRAYSCALE;
                    channels = 1;
                    break;
                case LOAD_COLOR:
                default:
                    type = CV_LOAD_IMAGE_COLOR;
                    channels = 3;
                    break;
            }

            byte[] px = new byte[channels];
            image = new Mat(height, width, type);
            ByteIndexer bi = image.createIndexer();


            for (int y=0; y<height; y++) {
                for (int x=0; x<width; x++) {
                    rgb = buf.getRGB(x, y);
                    px[0] = (byte)(rgb & 0xFF);
                    if (channels==3) {
                        px[1] = (byte)((rgb >> 8) & 0xFF);
                        px[2] = (byte)((rgb >> 16) & 0xFF);
                    }
                    bi.put(y, x, px);

                }
            }            
        }

        return image;

    }

    public static Mat readImage(String name) {
        return readImage(name, LOAD_COLOR);
    }

}

[saudet]

Using ImageIO isn't OpenCV, so I wouldn't call this class OpenCV anything. Also, FrameConverter already has more efficient methods to convert BufferedImage<->Mat.

In short, if you could refactor that to use FrameConverter, and then place these methods in say the JavaCV util class, I'll be happy to merge that! Thanks

[saudet]

Please send a pull request when you are ready to merge this! Thanks

[saudet]

Any updates on this?

[mavriksc]

very sorry about this. being more familiar with this now, let me try something over the weekend and if not able to get it done i'll close.

[mavriksc]

So it turns out i only used that method when i was running locally. if i built to jar and deployed web app it could not load the images from the class path with that method.
i ended up using this to get from file->Mat not sure if there is something that will already do file -> Frame i guess. :

public static Mat matFromJar(String path, int flags) {
        InputStream is = OpenCVUtils.class.getClassLoader().getResourceAsStream(path);
        if (is == null) {
                       System.out.println("ResourceNotFound: " + path);
                        return new Mat();
                }
        int nRead;
        byte[] data = new byte[16 * 1024];
        ByteArrayOutputStream buffer = new ByteArrayOutputStream();
        try {
            while ((nRead = is.read(data, 0, data.length)) != -1) {
                buffer.write(data, 0, nRead);
            }
        } catch (IOException e) {
            log.error("something happened",e);
        }
        byte[] bytes = buffer.toByteArray();
        Mat mat = imdecode(new Mat(bytes), flags);
        return mat;
    }

i also have some convenience methods

 public static Mat getROICopy(Mat m,Rect r){        
        Mat roi = new Mat(r.size(),m.type());
        m.rowRange(r.y(),r.y()+r.height()).colRange(r.x(), r.x()+r.width()).copyTo(roi);
        return roi;

    }

    public static Mat getROI(Mat m,Rect r){
              // Changes to the returned mat show on the original image     
        return m.rowRange(r.y(),r.y()+r.height()).colRange(r.x(), r.x()+r.width());

    }
        // m: the matching strength output from template matching
        // t: threshold value to count as a match
    public static List<Point> getPointsFromMatAboveThreshold(Mat m, float t){
        List<Point> matches = new ArrayList<Point>();
        FloatIndexer indexer = m.createIndexer();
        for (int y = 0; y < m.rows(); y++) {
            for (int x = 0; x < m.cols(); x++) {
                if (indexer.get(y,x)>t) {
                    System.out.println("(" + x + "," + y +") = "+ indexer.get(y,x));
                    matches.add(new Point(x, y));                   
                }
            }           
        }       
        return matches;
    }   
    // Given source image and template to match return max score
    public static float matchScore(Mat src,Mat tmp){

        Size size = new Size(src.cols()-tmp.cols()+1, src.rows()-tmp.rows()+1);
        Mat result = new Mat(size, CV_32FC1);
        matchTemplate(src, tmp, result, TM_CCORR_NORMED);
        DoublePointer minVal= new DoublePointer();
        DoublePointer maxVal= new DoublePointer();
        Point min = new Point();
        Point max = new Point();
        FloatIndexer fi = result.createIndexer();
        minMaxLoc(result, minVal, maxVal, min, max, null);      
        return fi.get(max.y(),max.x());
    } 

[saudet]

Looks useful! Could you put all that together and send a pull request to get this merged?