Camera recognition method for handwritten fonts

实现步骤：

1.正确配置android stdio工程，移植tensorflow和opencv

D:\文档\摄像头识别手写字体目录下已经准备好opencv调用摄像头文件的源码和布局文件，替换工程中的Main文件和xml文件

3.在androidManifest.xml文件中加入摄像头权限

4.摄像头权限动态申请

if (checkSelfPermission(Manifest.permission.CAMERA) != PackageManager.PERMISSION_GRANTED) { //申请权限 requestPermissions(new String[]{Manifest.permission.CAMERA}, 100); }

5.如遇手机显示问题以及黑屏闪退问题，需要手动调整分辨率

mOpenCvCameraView.setMaxFrameSize(640, 640);

把上一步的手写字体代码识别复制过来

private TensorFlowInferenceInterface tensorFlowInferenceInterface = null; private static final String mode_file = "file:///android_asset/MnistTF_model.pb"; private static final String INPUT_NODE = "conv2d_1_input_2:0"; private static final String OUTPUT_NODE = "dense_3_2/Softmax:0"; private float[] inputs_data = new float[784]; private float[] outputs_data = new float[10];

实例Tensorflow接口 tensorFlowInferenceInterface = new TensorFlowInferenceInterface(getAssets(), mode_file);

写一个函数用来识别手写字体

int width = bitmap_roi.getWidth(); int height = bitmap_roi.getHeight(); int[] pixels = new int[width * height];

Log.d("tag", width+" "+height);

try { bitmap_roi.getPixels(pixels, 0, width, 0, 0, width, height); for (int i = 0; i < pixels.length; i++) { inputs_data[i] = (float)pixels[i]; } }catch (Exception e){ Log.d("tag", e.getMessage()); }

Log.d("Tag", "width: "+width+" height:"+height);

Trace.beginSection("feed"); tensorFlowInferenceInterface.feed(INPUT_NODE, inputs_data, 1,28,28,1); Trace.endSection();

Trace.beginSection("run"); tensorFlowInferenceInterface.run(new String[]{OUTPUT_NODE}); Trace.endSection();

Trace.beginSection("fetch"); tensorFlowInferenceInterface.fetch(OUTPUT_NODE, outputs_data); Trace.endSection();

int logit = 0; for(int i=1;i<10;i++) { if(outputs_data[i]>outputs_data[logit]) logit=i; }

if(outputs_data[logit]>0) return logit; return -1;

我们的环境是白底黑字，接下来我们要用opencv把我们写的字从图片中分割出来

Mat img_t = new Mat(); Mat img_contours;

if(img_rgb != null){ Imgproc.cvtColor(img_rgb, img_gray, Imgproc.COLOR_RGB2GRAY);

Imgproc.threshold(img_gray, img_gray,140,255,Imgproc.THRESH_BINARY_INV);
Mat ele1=Imgproc.getStructuringElement(Imgproc.MORPH_RECT, new Size(3,3));
Mat ele2=Imgproc.getStructuringElement(Imgproc.MORPH_RECT, new Size(6,6));
Imgproc.erode(img_gray,img_gray,ele1);
Imgproc.dilate(img_gray,img_gray,ele2);

img_contours = img_gray.clone();
List<MatOfPoint> contours=new ArrayList<>();
Imgproc.findContours(img_contours,contours,new Mat(),
        Imgproc.RETR_LIST,Imgproc.CHAIN_APPROX_SIMPLE);
for (int contourIdx = 0; contourIdx < contours.size(); contourIdx++)
{
    double contourArea = Imgproc.contourArea(contours.get(contourIdx));
    Rect rect =  Imgproc.boundingRect(contours.get(contourIdx));
    if(contourArea<1500||contourArea>20000)
        continue;

    Mat roi = new Mat(img_gray, rect);
    Imgproc.resize(roi,roi,new Size(28,28));

    Bitmap bitmap2 = Bitmap.createBitmap(roi.width(),roi.height(),Bitmap.Config.RGB_565);
    Utils.matToBitmap(roi,bitmap2);
    int number = toNumber(bitmap2);
    if(number>=0) {
        //tl左上角顶点  br右下角定点
        double x = rect.tl().x;
        double y = rect.br().y;
        Point p = new Point(x,y);
        Imgproc.rectangle(img_rgb,rect.tl(),rect.br(), new Scalar(0,0,255));
        Imgproc.putText(img_rgb, Integer.toString(number), p, Core.FONT_HERSHEY_DUPLEX,
                6, new Scalar(0,0,255), 2);
    }
}

HaldenMemphis/Camera-recognition-method-for-handwritten-fonts

Camera recognition method for handwritten fonts