Java OpenCV FileStorage 和 Mat.push_back

Question

我正在尝试用 Java 为 KNN 分类器实现 this 项目，即 GenData.cpp（用 C++ 编写）。
我已经到达这些代码行并卡住了：

matClassificationInts.push_back(intChar);
cv::FileStorage fsClassifications("classifications.xml", cv::FileStorage::WRITE);
fsClassifications << "classifications" << matClassificationInts;
fsClassifications.release();

在 c++ 中，我们可以将整数传递给 push_back()，但在 Java 中我收到错误：“int cannot be convert to Mat”。
所以，第一个问题是：如何将 int 传递给 someMat.push_back()？
第二个：如何在 Java 中实现 FileStorage 或将 Mat 写入 *.xml 格式（并从 *.xml 读取 Mat）？

到目前为止，我的代码：

    import java.io.IOException;
    import java.util.ArrayList;
    import java.util.Arrays;
    import java.util.Scanner;
    import org.opencv.core.Core;
    import static org.opencv.core.CvType.CV_32FC1;
    import org.opencv.core.Mat;
    import org.opencv.core.MatOfInt4;
    import org.opencv.core.MatOfPoint;
    import org.opencv.core.Rect;
    import org.opencv.core.Scalar;
    import org.opencv.core.Size;
    import org.opencv.imgcodecs.Imgcodecs;
    import org.opencv.imgproc.Imgproc;
    import static org.opencv.imgproc.Imgproc.ADAPTIVE_THRESH_GAUSSIAN_C;
    import static org.opencv.imgproc.Imgproc.CHAIN_APPROX_SIMPLE;
    import static org.opencv.imgproc.Imgproc.RETR_EXTERNAL;
    import static org.opencv.imgproc.Imgproc.THRESH_BINARY_INV;

    public class genData {

    private static final int 
            MIN_CONTOUR_AREA = 100,
            RESIZED_IMAGE_WIDTH = 20,
            RESIZED_IMAGE_HEIGHT = 30;

    public static void main(String[] args) throws IOException {
        System.loadLibrary(Core.NATIVE_LIBRARY_NAME); 

        Scanner keyboard = new Scanner(System.in);
        boolean exit = false;

        Mat imgTrainingNumbers;
        Mat imgGrayscale = new Mat();
        Mat imgBlurred = new Mat();
        Mat imgThresh = new Mat();
        Mat imgThreshCopy = new Mat();

        ArrayList<MatOfPoint> ptContours = new ArrayList<MatOfPoint>();
        MatOfInt4 v4iHierarchy;
        Mat matClassificationInts = new Mat();
        Mat matTrainingImagesAsFlattenedFloats = new Mat();

        int[] intValidChars = { '0', '1', '2',
        'A', 'B', 'C'}; //Here I did not make List<Integer>, because I can't pass char to Integer.
        Arrays.sort(intValidChars); //for binary search

        imgTrainingNumbers = Imgcodecs.imread("test.png"); //here Text on white paper.

        if (imgTrainingNumbers.empty()) {
            System.out.println("err");
            return;
        }

        Imgproc.cvtColor(imgTrainingNumbers, imgGrayscale, Imgproc.COLOR_BGR2GRAY);
        Imgproc.GaussianBlur(imgGrayscale, imgBlurred, new Size(5, 5), 0);
        Imgproc.adaptiveThreshold(imgBlurred, imgThresh, 255, ADAPTIVE_THRESH_GAUSSIAN_C, THRESH_BINARY_INV, 11, 2);

        /*
        //imshow class implementation (found via google, works properly, but this block is commented for now)
        Imshow im = new Imshow("imgThresh");
        im.showImage(imgThresh);
        imgThreshCopy = imgThresh.clone();
        */

        Imgproc.findContours(imgThreshCopy, ptContours, new Mat(), RETR_EXTERNAL, CHAIN_APPROX_SIMPLE);


        for (int i = 0; i < ptContours.size(); i++) {
            if (Imgproc.contourArea(ptContours.get(i)) > MIN_CONTOUR_AREA) {
                Rect boundingRect = Imgproc.boundingRect(ptContours.get(i));
                Imgproc.rectangle(imgTrainingNumbers, boundingRect.tl(), boundingRect.br(), new Scalar(0, 0, 255), 2);
                Mat matROI = imgThresh.submat(boundingRect.y, boundingRect.y + boundingRect.height, boundingRect.x, boundingRect.x + boundingRect.width);
                Mat matROIResized = new Mat();
                Imgproc.resize(matROI, matROIResized, new Size(RESIZED_IMAGE_WIDTH, RESIZED_IMAGE_HEIGHT));
                /*
                im.showImage(matROI);
                im.showImage(matROIResized);
                im.showImage(imgTrainingNumbers);
                */
                String input = keyboard.nextLine();
                int intChar = (int)input.charAt(0);
                if (Arrays.binarySearch(intValidChars, intChar) >=0) {
                    /*
                    matClassificationInts.push_back(intChar);
                    //Here I'm getting an error.
                    */
                    Mat matImageFloat = new Mat();
                    matROIResized.convertTo(matImageFloat, CV_32FC1);
                    Mat matImageFlattenedFloat = matImageFloat.reshape(1, 1);
                    matTrainingImagesAsFlattenedFloats.push_back(matImageFlattenedFloat);
                }
            }
        }
       //Here should go FileStorage stuff.
    }   
}

提前致谢。
附：使用 OpenCV_310 + Java（不是 JavaCV）

Answer 1

好吧，我已经自己解决了。 很脏，我猜，但就是这样。如果您知道如何对我的代码进行良好的改进，我会很高兴阅读您的 cmets。
1）我的第一个问题是关于将 int 放入 Mat （用于进一步制作 *.xml）。我避免了这种方法，并决定将 int（实际上是 Integer）放入 List。

 Scanner keyboard = new Scanner(System.in);
 String input = keyboard.nextLine();
 int intChar = (int)input.charAt(0);
 List<Integer> matClassificationInts = new ArrayList<Integer>();
 if (Arrays.binarySearch(intValidChars, intChar) >=0) {
    matClassificationInts.add(new Integer(intChar));
    ......
 }
 String dataImages = "";
 for (Integer i : matClassificationInts) {
     dataImages += i + " ";
 }

我可以制作字符串（例如“49 48”字符“1 0”整数）以将其保存在 *.xml 中（参见下一段）。
2) 第二个问题是关于从 Mat 中提取数据并将其存储在 *.xml 中。好吧，通过 C++，我可以通过 FileStorage 来实现：

cv::FileStorage fsClassifications("classifications.xml", cv::FileStorage::WRITE);
fsClassifications << "classifications" << matClassificationInts;
fsClassifications.release();

但是 Java OpenCV 没有这样的功能，所以我循环遍历二维数组（Mat.rows() 和 Mat.cols()）并通过 get() 方法提取所需的数据（Mat.get(row, col) -给出双精度数组，数组长度 = 1)：

String dataClassifications = "";
for (int i = 0; i < matTrainingImagesAsFlattenedFloats.rows(); i++) {
    for (int j = 0; j < matTrainingImagesAsFlattenedFloats.cols(); j++) {
        double[] temp = matTrainingImagesAsFlattenedFloats.get(i, j);
        dataClassifications += temp[0] + " ";
    }
    dataClassifications += "\n";
}

现在，关于将数据保存到 *.xml：
我刚刚使用了 javafx.xml 和 org.wc3.dom 库。
制作了两个用于返回 DOM 节点的函数：

private static Node getMatXML(Document doc, String option_id, String type_id, String rows, String cols, String dt, String data) {
    Element elem = doc.createElement(option_id);
    elem.setAttribute("type_id", type_id);
    elem.appendChild(getMatXMLElement(doc,"rows", rows));
    elem.appendChild(getMatXMLElement(doc, "cols", cols));
    elem.appendChild(getMatXMLElement(doc, "dt", dt));
    elem.appendChild(getMatXMLElement(doc, "data", data));
    return elem;
}

private static Node getMatXMLElement(Document doc, String name, String value) {
    Element node = doc.createElement(name);
    node.appendChild(doc.createTextNode(value));
    return node;
}

并使用这些函数来创建 *.xml:
分类.xml：

    DocumentBuilderFactory icFactory_images = DocumentBuilderFactory.newInstance();
    DocumentBuilder icBuilder_images;
    try {
        icBuilder_images = icFactory_images.newDocumentBuilder();
        Document doc = icBuilder_images.newDocument();
        Element mainRootElement = doc.createElement("opencv_storage");
        doc.appendChild(mainRootElement);
        mainRootElement.appendChild(getMatXML(doc, "classifications", "opencv-matrix", rowsImages, colsImages, "i", dataImages));
        Transformer transformer = TransformerFactory.newInstance().newTransformer();
        transformer.setOutputProperty(OutputKeys.INDENT, "yes"); 
        DOMSource source = new DOMSource(doc);
        String filename = "classifications.xml";
        File file = new File(filename);
        StreamResult console = new StreamResult(file); //(System.out)
        transformer.transform(source, console);
    } catch (Exception e) {
        e.printStackTrace();
    }

Images.xml：

    DocumentBuilderFactory icFactory_classifications = DocumentBuilderFactory.newInstance();
    DocumentBuilder icBuilder_classifications;
    try {
        icBuilder_classifications = icFactory_classifications.newDocumentBuilder();
        Document doc = icBuilder_classifications.newDocument();
        Element mainRootElement = doc.createElement("opencv_storage");
        doc.appendChild(mainRootElement);
        mainRootElement.appendChild(getMatXML(doc, "images", "opencv-matrix", rowsClassifications, colsClassifications, "f", dataClassifications));
        Transformer transformer = TransformerFactory.newInstance().newTransformer();
        transformer.setOutputProperty(OutputKeys.INDENT, "yes"); 
        DOMSource source = new DOMSource(doc);
        String filename = "images.xml";
        File file = new File(filename);
        StreamResult console = new StreamResult(file); //(System.out)
        transformer.transform(source, console);
    } catch (Exception e) {
        e.printStackTrace();
    }

例如，生成的分类文件是：

<?xml version="1.0" encoding="UTF-8" standalone="no"?>
<opencv_storage>
<classifications type_id="opencv-matrix">
<rows>2</rows>
<cols>1</cols>
<dt>i</dt>
<data>49 48 </data>
</classifications>
</opencv_storage>

我为这张照片做了测试：

通过 GenData.cpp（参见相关链接 - 第 1 行）和我的 Java 代码（完整代码见下文）。两个程序都给了我相同的结果：
对于 Java OpenCV Imshow 实现，您可以查看 this link（不是我的）。

import java.io.File;
import java.io.FileNotFoundException;
import java.io.FileOutputStream;
import java.io.FileReader;
import java.io.IOException;
import java.util.ArrayList;
import java.util.Arrays;
import java.util.List;
import java.util.Map;
import java.util.Scanner;
import org.opencv.core.Core;
import static org.opencv.core.CvType.CV_32FC1;
import org.opencv.core.Mat;
import org.opencv.core.MatOfInt4;
import org.opencv.core.MatOfPoint;
import org.opencv.core.Rect;
import org.opencv.core.Scalar;
import org.opencv.core.Size;
import org.opencv.imgcodecs.Imgcodecs;
import org.opencv.imgproc.Imgproc;
import static org.opencv.imgproc.Imgproc.ADAPTIVE_THRESH_GAUSSIAN_C;
import static org.opencv.imgproc.Imgproc.CHAIN_APPROX_SIMPLE;
import static org.opencv.imgproc.Imgproc.RETR_EXTERNAL;
import static org.opencv.imgproc.Imgproc.THRESH_BINARY_INV;

//XML - write.
import javax.xml.parsers.DocumentBuilder;
import javax.xml.parsers.DocumentBuilderFactory;
import javax.xml.transform.OutputKeys;
import javax.xml.transform.Transformer;
import javax.xml.transform.TransformerFactory;
import javax.xml.transform.dom.DOMSource;
import javax.xml.transform.stream.StreamResult;
import org.w3c.dom.Document;
import org.w3c.dom.Element;
import org.w3c.dom.Node;

public class genData {


    private static final int 
            MIN_CONTOUR_AREA = 100,
            RESIZED_IMAGE_WIDTH = 20,
            RESIZED_IMAGE_HEIGHT = 30;

    public static void main(String[] args) throws IOException {
        System.loadLibrary(Core.NATIVE_LIBRARY_NAME); 
        Scanner keyboard = new Scanner(System.in);

        Mat imgTrainingNumbers;
        Mat imgGrayscale = new Mat();
        Mat imgBlurred = new Mat();
        Mat imgThresh = new Mat();
        Mat imgThreshCopy = new Mat();

        ArrayList<MatOfPoint> ptContours = new ArrayList<MatOfPoint>();
        MatOfInt4 v4iHierarchy = new MatOfInt4();

        List<Integer> matClassificationInts = new ArrayList<Integer>();

        Mat matTrainingImagesAsFlattenedFloats = new Mat();

        int[] intValidChars = { '0', '1', '2', '3', '4', '5', '6', '7', '8', '9',
        'A', 'B', 'C', 'E', 'H',
        'K', 'M', 'O', 'P', 'T',
        'X', 'Y'};
        Arrays.sort(intValidChars);

        imgTrainingNumbers = Imgcodecs.imread("01.png");

        if (imgTrainingNumbers.empty()) {
            System.out.println("Error: file is not found");
            return;
        }

        Imgproc.cvtColor(imgTrainingNumbers, imgGrayscale, Imgproc.COLOR_BGR2GRAY);
        Imgproc.GaussianBlur(imgGrayscale, imgBlurred, new Size(5, 5), 0);
        Imgproc.adaptiveThreshold(imgBlurred, imgThresh, 255, ADAPTIVE_THRESH_GAUSSIAN_C, THRESH_BINARY_INV, 11, 2);

        Imshow im = new Imshow("imgThresh");
        im.showImage(imgThresh);
        imgThreshCopy = imgThresh.clone();

        Imgproc.findContours(imgThreshCopy, ptContours, v4iHierarchy, RETR_EXTERNAL, CHAIN_APPROX_SIMPLE);


        for (int i = 0; i < ptContours.size(); i++) {
            if (Imgproc.contourArea(ptContours.get(i)) > MIN_CONTOUR_AREA) {
                Rect boundingRect = Imgproc.boundingRect(ptContours.get(i));
                Imgproc.rectangle(imgTrainingNumbers, boundingRect.tl(), boundingRect.br(), new Scalar(0, 0, 255), 2);
                Mat matROI = imgThresh.submat(boundingRect.y, boundingRect.y + boundingRect.height, boundingRect.x, boundingRect.x + boundingRect.width);
                Mat matROIResized = new Mat();
                Imgproc.resize(matROI, matROIResized, new Size(RESIZED_IMAGE_WIDTH, RESIZED_IMAGE_HEIGHT));
                im.showImage(matROI);
                im.showImage(matROIResized);
                im.showImage(imgTrainingNumbers);
                String input = keyboard.nextLine();
                int intChar = (int)input.charAt(0);
                if (Arrays.binarySearch(intValidChars, intChar) >=0) {
                    matClassificationInts.add(new Integer(intChar));
                    Mat matImageFloat = new Mat();
                    matROIResized.convertTo(matImageFloat, CV_32FC1);
                    Mat matImageFlattenedFloat = matImageFloat.reshape(1, 1);
                    matTrainingImagesAsFlattenedFloats.push_back(matImageFlattenedFloat);
                }
            }
        }

        String dataImages = "";
        for (Integer i : matClassificationInts) {
            dataImages += i + " ";
        }

        String dataClassifications = "";
        for (int i = 0; i < matTrainingImagesAsFlattenedFloats.rows(); i++) {
            for (int j = 0; j < matTrainingImagesAsFlattenedFloats.cols(); j++) {
                double[] temp = matTrainingImagesAsFlattenedFloats.get(i, j);
                dataClassifications += temp[0] + " ";
            }
            dataClassifications += "\n";
        }

        String rowsImages = String.valueOf(matClassificationInts.size());
        String colsImages = "1";
        String rowsClassifications = String.valueOf(matTrainingImagesAsFlattenedFloats.rows());
        String colsClassifications = String.valueOf(matTrainingImagesAsFlattenedFloats.cols());

        DocumentBuilderFactory icFactory_images = DocumentBuilderFactory.newInstance();
        DocumentBuilder icBuilder_images;
        try {
            icBuilder_images = icFactory_images.newDocumentBuilder();
            Document doc = icBuilder_images.newDocument();
            Element mainRootElement = doc.createElement("opencv_storage");
            doc.appendChild(mainRootElement);
            mainRootElement.appendChild(getMatXML(doc, "classifications", "opencv-matrix", rowsImages, colsImages, "i", dataImages));
            Transformer transformer = TransformerFactory.newInstance().newTransformer();
            transformer.setOutputProperty(OutputKeys.INDENT, "yes"); 
            DOMSource source = new DOMSource(doc);
            String filename = "classifications.xml";
            File file = new File(filename);
            StreamResult console = new StreamResult(file); //(System.out)
            transformer.transform(source, console);
        } catch (Exception e) {
            e.printStackTrace();
        }

        DocumentBuilderFactory icFactory_classifications = DocumentBuilderFactory.newInstance();
        DocumentBuilder icBuilder_classifications;
        try {
            icBuilder_classifications = icFactory_classifications.newDocumentBuilder();
            Document doc = icBuilder_classifications.newDocument();
            Element mainRootElement = doc.createElement("opencv_storage");
            doc.appendChild(mainRootElement);
            mainRootElement.appendChild(getMatXML(doc, "images", "opencv-matrix", rowsClassifications, colsClassifications, "f", dataClassifications));
            Transformer transformer = TransformerFactory.newInstance().newTransformer();
            transformer.setOutputProperty(OutputKeys.INDENT, "yes"); 
            DOMSource source = new DOMSource(doc);
            String filename = "images.xml";
            File file = new File(filename);
            StreamResult console = new StreamResult(file); //(System.out)
            transformer.transform(source, console);
        } catch (Exception e) {
            e.printStackTrace();
        }
        System.out.println("Finished.");
        System.exit(0);
    }   

    private static Node getMatXML(Document doc, String option_id, String type_id, String rows, String cols, String dt, String data) {
        Element elem = doc.createElement(option_id);
        elem.setAttribute("type_id", type_id);
        elem.appendChild(getMatXMLElement(doc,"rows", rows));
        elem.appendChild(getMatXMLElement(doc, "cols", cols));
        elem.appendChild(getMatXMLElement(doc, "dt", dt));
        elem.appendChild(getMatXMLElement(doc, "data", data));
        return elem;
    }

    private static Node getMatXMLElement(Document doc, String name, String value) {
        Element node = doc.createElement(name);
        node.appendChild(doc.createTextNode(value));
        return node;
    }
}