Helps in detecting text characters on metal plates

Android/Java
1

Hi everyone, I’m new to OpenCV.
I’m trying to identify text on metal plates and vehicle chassis.

This would be the desired effect:

placa
placa1229×753 211 KB

Tried this implementation below:

package com.autoscore;

import nu.pattern.OpenCV;
import org.opencv.core.*;
import org.opencv.highgui.HighGui;
import org.opencv.imgcodecs.Imgcodecs;
import org.opencv.imgproc.Imgproc;

import java.util.ArrayList;
import java.util.List;

public class CharDetect {

    public static void main(String[] args) {
        // Carregue a biblioteca OpenCV
        OpenCV.loadLocally();

        // Load image, grayscale, Otsu's threshold
        String imagePath = "./src/main/resources/chassi3.jpeg";

        // Carrega a imagem
        Mat image = Imgcodecs.imread(imagePath, Imgcodecs.IMREAD_GRAYSCALE);

        // Aplica binarização adaptativa para melhorar a detecção de texto branco em fundo escuro
        Mat adaptiveThreshold = new Mat();
        Imgproc.adaptiveThreshold(image, adaptiveThreshold, 255, Imgproc.ADAPTIVE_THRESH_GAUSSIAN_C, Imgproc.THRESH_BINARY_INV, 11, 3);

        // Aplica um filtro de média para remover ruídos
        Mat filteredImage = new Mat();
        Imgproc.medianBlur(adaptiveThreshold, filteredImage, 5);

        // Encontra os contornos na imagem filtrada
        Mat hierarchy = new Mat();
        List<MatOfPoint> contours = new ArrayList<>();
        Imgproc.findContours(filteredImage, contours, hierarchy, Imgproc.RETR_EXTERNAL, Imgproc.CHAIN_APPROX_SIMPLE);

        // Desenha os contornos na imagem original
        Mat contourImage = new Mat();
        Imgproc.cvtColor(image, contourImage, Imgproc.COLOR_GRAY2BGR);
        Scalar color = new Scalar(0, 0, 255); // Cor vermelha para os contornos
        int thickness = 2; // Espessura da linha do contorno

        for (MatOfPoint contour : contours) {
            Rect rect = Imgproc.boundingRect(contour);
            double aspectRatio = (double) rect.width / rect.height;

            // Verifica se a proporção da largura pela altura está próxima à proporção de um caractere de texto comum
            if (aspectRatio > 0.1 && aspectRatio < 1.0) {
                double area = Imgproc.contourArea(contour);
                if (area > 100) { // Ajuste o valor do limiar conforme necessário para evitar contornos indesejados
                    Imgproc.rectangle(contourImage, rect.tl(), rect.br(), color, thickness);
                }
            }
        }


        HighGui.imshow("thresh", contourImage);
       // HighGui.imshow("croppedImage", croppedImage);

        HighGui.waitKey();
    }
}

But this was the only result I got:

798×155 65.6 KB

2

you could try to exploit, that this is a monospaced font, and you mainly need to detect ‘a grid of letters’
(they’re all same width / height !)

3

Would you have any examples of how to do this, I’m really lost.

4

these situations are usually solved by training a neural network.

no, I highly doubt that tesseract ocr can solve any non-trivial tasks. don’t even think about that. use something else.

5

Could you provide a sample original image

6

IMG_20220611_115619557_HDR