Extracting text OpenCV

Question

I am trying to find the bounding boxes of text in an image and am currently using this approach:

// calculate the local variances of the grayscale image
Mat          


        

Answer 1

This is a C# version of the answer from dhanushka using OpenCVSharp

        Mat large = new Mat(INPUT_FILE);
        Mat rgb = new Mat(), small = new Mat(), grad = new Mat(), bw = new Mat(), connected = new Mat();

        // downsample and use it for processing
        Cv2.PyrDown(large, rgb);
        Cv2.CvtColor(rgb, small, ColorConversionCodes.BGR2GRAY);

        // morphological gradient
        var morphKernel = Cv2.GetStructuringElement(MorphShapes.Ellipse, new OpenCvSharp.Size(3, 3));
        Cv2.MorphologyEx(small, grad, MorphTypes.Gradient, morphKernel);

        // binarize
        Cv2.Threshold(grad, bw, 0, 255, ThresholdTypes.Binary | ThresholdTypes.Otsu);

        // connect horizontally oriented regions
        morphKernel = Cv2.GetStructuringElement(MorphShapes.Rect, new OpenCvSharp.Size(9, 1));
        Cv2.MorphologyEx(bw, connected, MorphTypes.Close, morphKernel);

        // find contours
        var mask = new Mat(Mat.Zeros(bw.Size(), MatType.CV_8UC1), Range.All);
        Cv2.FindContours(connected, out OpenCvSharp.Point[][] contours, out HierarchyIndex[] hierarchy, RetrievalModes.CComp, ContourApproximationModes.ApproxSimple, new OpenCvSharp.Point(0, 0));

        // filter contours
        var idx = 0;
        foreach (var hierarchyItem in hierarchy)
        {
            idx = hierarchyItem.Next;
            if (idx < 0)
                break;
            OpenCvSharp.Rect rect = Cv2.BoundingRect(contours[idx]);
            var maskROI = new Mat(mask, rect);
            maskROI.SetTo(new Scalar(0, 0, 0));

            // fill the contour
            Cv2.DrawContours(mask, contours, idx, Scalar.White, -1);

            // ratio of non-zero pixels in the filled region
            double r = (double)Cv2.CountNonZero(maskROI) / (rect.Width * rect.Height);
            if (r > .45 /* assume at least 45% of the area is filled if it contains text */
                 &&
            (rect.Height > 8 && rect.Width > 8) /* constraints on region size */
            /* these two conditions alone are not very robust. better to use something 
            like the number of significant peaks in a horizontal projection as a third condition */
            )
            {
                Cv2.Rectangle(rgb, rect, new Scalar(0, 255, 0), 2);
            }
        }

        rgb.SaveImage(Path.Combine(AppDomain.CurrentDomain.BaseDirectory, "rgb.jpg"));

Answer 2

this is a VB.NET version of the answer from dhanushka using EmguCV.

A few functions and structures in EmguCV need different consideration than the C# version with OpenCVSharp

Imports Emgu.CV
Imports Emgu.CV.Structure
Imports Emgu.CV.CvEnum
Imports Emgu.CV.Util

        Dim input_file As String = "C:\your_input_image.png"
        Dim large As Mat = New Mat(input_file)
        Dim rgb As New Mat
        Dim small As New Mat
        Dim grad As New Mat
        Dim bw As New Mat
        Dim connected As New Mat
        Dim morphanchor As New Point(0, 0)

        '//downsample and use it for processing
        CvInvoke.PyrDown(large, rgb)
        CvInvoke.CvtColor(rgb, small, ColorConversion.Bgr2Gray)

        '//morphological gradient
        Dim morphKernel As Mat = CvInvoke.GetStructuringElement(ElementShape.Ellipse, New Size(3, 3), morphanchor)
        CvInvoke.MorphologyEx(small, grad, MorphOp.Gradient, morphKernel, New Point(0, 0), 1, BorderType.Isolated, New MCvScalar(0))

        '// binarize
        CvInvoke.Threshold(grad, bw, 0, 255, ThresholdType.Binary Or ThresholdType.Otsu)

        '// connect horizontally oriented regions
        morphKernel = CvInvoke.GetStructuringElement(ElementShape.Rectangle, New Size(9, 1), morphanchor)
        CvInvoke.MorphologyEx(bw, connected, MorphOp.Close, morphKernel, morphanchor, 1, BorderType.Isolated, New MCvScalar(0))

        '// find contours
        Dim mask As Mat = Mat.Zeros(bw.Size.Height, bw.Size.Width, DepthType.Cv8U, 1)  '' MatType.CV_8UC1
        Dim contours As New VectorOfVectorOfPoint
        Dim hierarchy As New Mat

        CvInvoke.FindContours(connected, contours, hierarchy, RetrType.Ccomp, ChainApproxMethod.ChainApproxSimple, Nothing)

        '// filter contours
        Dim idx As Integer
        Dim rect As Rectangle
        Dim maskROI As Mat
        Dim r As Double
        For Each hierarchyItem In hierarchy.GetData
            rect = CvInvoke.BoundingRectangle(contours(idx))
            maskROI = New Mat(mask, rect)
            maskROI.SetTo(New MCvScalar(0, 0, 0))

            '// fill the contour
            CvInvoke.DrawContours(mask, contours, idx, New MCvScalar(255), -1)

            '// ratio of non-zero pixels in the filled region
            r = CvInvoke.CountNonZero(maskROI) / (rect.Width * rect.Height)

            '/* assume at least 45% of the area Is filled if it contains text */
            '/* constraints on region size */
            '/* these two conditions alone are Not very robust. better to use something 
            'Like the number of significant peaks in a horizontal projection as a third condition */
            If r > 0.45 AndAlso rect.Height > 8 AndAlso rect.Width > 8 Then
                'draw green rectangle
                CvInvoke.Rectangle(rgb, rect, New MCvScalar(0, 255, 0), 2)
            End If
            idx += 1
        Next
        rgb.Save(IO.Path.Combine(Application.StartupPath, "rgb.jpg"))

Answer 3

I used a gradient based method in the program below. Added the resulting images. Please note that I'm using a scaled down version of the image for processing.

c++ version

The MIT License (MIT)

Copyright (c) 2014 Dhanushka Dangampola

Permission is hereby granted, free of charge, to any person obtaining a copy
of this software and associated documentation files (the "Software"), to deal
in the Software without restriction, including without limitation the rights
to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
copies of the Software, and to permit persons to whom the Software is
furnished to do so, subject to the following conditions:

The above copyright notice and this permission notice shall be included in
all copies or substantial portions of the Software.

THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN
THE SOFTWARE.

#include "stdafx.h"

#include <opencv2/core/core.hpp>
#include <opencv2/highgui/highgui.hpp>
#include <opencv2/imgproc/imgproc.hpp>
#include <iostream>

using namespace cv;
using namespace std;

#define INPUT_FILE              "1.jpg"
#define OUTPUT_FOLDER_PATH      string("")

int _tmain(int argc, _TCHAR* argv[])
{
    Mat large = imread(INPUT_FILE);
    Mat rgb;
    // downsample and use it for processing
    pyrDown(large, rgb);
    Mat small;
    cvtColor(rgb, small, CV_BGR2GRAY);
    // morphological gradient
    Mat grad;
    Mat morphKernel = getStructuringElement(MORPH_ELLIPSE, Size(3, 3));
    morphologyEx(small, grad, MORPH_GRADIENT, morphKernel);
    // binarize
    Mat bw;
    threshold(grad, bw, 0.0, 255.0, THRESH_BINARY | THRESH_OTSU);
    // connect horizontally oriented regions
    Mat connected;
    morphKernel = getStructuringElement(MORPH_RECT, Size(9, 1));
    morphologyEx(bw, connected, MORPH_CLOSE, morphKernel);
    // find contours
    Mat mask = Mat::zeros(bw.size(), CV_8UC1);
    vector<vector<Point>> contours;
    vector<Vec4i> hierarchy;
    findContours(connected, contours, hierarchy, CV_RETR_CCOMP, CV_CHAIN_APPROX_SIMPLE, Point(0, 0));
    // filter contours
    for(int idx = 0; idx >= 0; idx = hierarchy[idx][0])
    {
        Rect rect = boundingRect(contours[idx]);
        Mat maskROI(mask, rect);
        maskROI = Scalar(0, 0, 0);
        // fill the contour
        drawContours(mask, contours, idx, Scalar(255, 255, 255), CV_FILLED);
        // ratio of non-zero pixels in the filled region
        double r = (double)countNonZero(maskROI)/(rect.width*rect.height);

        if (r > .45 /* assume at least 45% of the area is filled if it contains text */
            && 
            (rect.height > 8 && rect.width > 8) /* constraints on region size */
            /* these two conditions alone are not very robust. better to use something 
            like the number of significant peaks in a horizontal projection as a third condition */
            )
        {
            rectangle(rgb, rect, Scalar(0, 255, 0), 2);
        }
    }
    imwrite(OUTPUT_FOLDER_PATH + string("rgb.jpg"), rgb);

    return 0;
}

python version

The MIT License (MIT)

Copyright (c) 2017 Dhanushka Dangampola

Permission is hereby granted, free of charge, to any person obtaining a copy
of this software and associated documentation files (the "Software"), to deal
in the Software without restriction, including without limitation the rights
to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
copies of the Software, and to permit persons to whom the Software is
furnished to do so, subject to the following conditions:

The above copyright notice and this permission notice shall be included in
all copies or substantial portions of the Software.

THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN
THE SOFTWARE.

import cv2
import numpy as np

large = cv2.imread('1.jpg')
rgb = cv2.pyrDown(large)
small = cv2.cvtColor(rgb, cv2.COLOR_BGR2GRAY)

kernel = cv2.getStructuringElement(cv2.MORPH_ELLIPSE, (3, 3))
grad = cv2.morphologyEx(small, cv2.MORPH_GRADIENT, kernel)

_, bw = cv2.threshold(grad, 0.0, 255.0, cv2.THRESH_BINARY | cv2.THRESH_OTSU)

kernel = cv2.getStructuringElement(cv2.MORPH_RECT, (9, 1))
connected = cv2.morphologyEx(bw, cv2.MORPH_CLOSE, kernel)
# using RETR_EXTERNAL instead of RETR_CCOMP
contours, hierarchy = cv2.findContours(connected.copy(), cv2.RETR_EXTERNAL, cv2.CHAIN_APPROX_NONE)
#For opencv 3+ comment the previous line and uncomment the following line
#_, contours, hierarchy = cv2.findContours(connected.copy(), cv2.RETR_EXTERNAL, cv2.CHAIN_APPROX_NONE)

mask = np.zeros(bw.shape, dtype=np.uint8)

for idx in range(len(contours)):
    x, y, w, h = cv2.boundingRect(contours[idx])
    mask[y:y+h, x:x+w] = 0
    cv2.drawContours(mask, contours, idx, (255, 255, 255), -1)
    r = float(cv2.countNonZero(mask[y:y+h, x:x+w])) / (w * h)

    if r > 0.45 and w > 8 and h > 8:
        cv2.rectangle(rgb, (x, y), (x+w-1, y+h-1), (0, 255, 0), 2)

cv2.imshow('rects', rgb)

Answer 4

You can detect text by finding close edge elements (inspired from a LPD):

#include "opencv2/opencv.hpp"

std::vector<cv::Rect> detectLetters(cv::Mat img)
{
    std::vector<cv::Rect> boundRect;
    cv::Mat img_gray, img_sobel, img_threshold, element;
    cvtColor(img, img_gray, CV_BGR2GRAY);
    cv::Sobel(img_gray, img_sobel, CV_8U, 1, 0, 3, 1, 0, cv::BORDER_DEFAULT);
    cv::threshold(img_sobel, img_threshold, 0, 255, CV_THRESH_OTSU+CV_THRESH_BINARY);
    element = getStructuringElement(cv::MORPH_RECT, cv::Size(17, 3) );
    cv::morphologyEx(img_threshold, img_threshold, CV_MOP_CLOSE, element); //Does the trick
    std::vector< std::vector< cv::Point> > contours;
    cv::findContours(img_threshold, contours, 0, 1); 
    std::vector<std::vector<cv::Point> > contours_poly( contours.size() );
    for( int i = 0; i < contours.size(); i++ )
        if (contours[i].size()>100)
        { 
            cv::approxPolyDP( cv::Mat(contours[i]), contours_poly[i], 3, true );
            cv::Rect appRect( boundingRect( cv::Mat(contours_poly[i]) ));
            if (appRect.width>appRect.height) 
                boundRect.push_back(appRect);
        }
    return boundRect;
}

Usage:

int main(int argc,char** argv)
{
    //Read
    cv::Mat img1=cv::imread("side_1.jpg");
    cv::Mat img2=cv::imread("side_2.jpg");
    //Detect
    std::vector<cv::Rect> letterBBoxes1=detectLetters(img1);
    std::vector<cv::Rect> letterBBoxes2=detectLetters(img2);
    //Display
    for(int i=0; i< letterBBoxes1.size(); i++)
        cv::rectangle(img1,letterBBoxes1[i],cv::Scalar(0,255,0),3,8,0);
    cv::imwrite( "imgOut1.jpg", img1);  
    for(int i=0; i< letterBBoxes2.size(); i++)
        cv::rectangle(img2,letterBBoxes2[i],cv::Scalar(0,255,0),3,8,0);
    cv::imwrite( "imgOut2.jpg", img2);  
    return 0;
}

Results:

a. element = getStructuringElement(cv::MORPH_RECT, cv::Size(17, 3) );

b. element = getStructuringElement(cv::MORPH_RECT, cv::Size(30, 30) );

Results are similar for the other image mentioned.

Answer 5

Python Implementation for @dhanushka's solution:

def process_rgb(rgb):
    hasText = False
    gray = cv2.cvtColor(rgb, cv2.COLOR_BGR2GRAY)
    morphKernel = cv2.getStructuringElement(cv2.MORPH_ELLIPSE, (3,3))
    grad = cv2.morphologyEx(gray, cv2.MORPH_GRADIENT, morphKernel)
    # binarize
    _, bw = cv2.threshold(grad, 0.0, 255.0, cv2.THRESH_BINARY | cv2.THRESH_OTSU)
    # connect horizontally oriented regions
    morphKernel = cv2.getStructuringElement(cv2.MORPH_RECT, (9, 1))
    connected = cv2.morphologyEx(bw, cv2.MORPH_CLOSE, morphKernel)
    # find contours
    mask = np.zeros(bw.shape[:2], dtype="uint8")
    _,contours, hierarchy = cv2.findContours(connected, cv2.RETR_CCOMP, cv2.CHAIN_APPROX_SIMPLE)
    # filter contours
    idx = 0
    while idx >= 0:
        x,y,w,h = cv2.boundingRect(contours[idx])
        # fill the contour
        cv2.drawContours(mask, contours, idx, (255, 255, 255), cv2.FILLED)
        # ratio of non-zero pixels in the filled region
        r = cv2.contourArea(contours[idx])/(w*h)
        if(r > 0.45 and h > 5 and w > 5 and w > h):
            cv2.rectangle(rgb, (x,y), (x+w,y+h), (0, 255, 0), 2)
            hasText = True
        idx = hierarchy[0][idx][0]
    return hasText, rgb

Answer 6

You can utilize a python implementation SWTloc.

Full Disclosure : I am the author of this library

To do that :-

First and Second Image

Notice that the text_mode here is 'lb_df', which stands for Light Background Dark Foreground i.e the text in this image is going to be in darker color than the background

from swtloc import SWTLocalizer
from swtloc.utils import imgshowN, imgshow

swtl = SWTLocalizer()
# Stroke Width Transform
swtl.swttransform(imgpaths='img1.jpg', text_mode = 'lb_df',
                  save_results=True, save_rootpath = 'swtres/',
                  minrsw = 3, maxrsw = 20, max_angledev = np.pi/3)
imgshow(swtl.swtlabelled_pruned13C)

# Grouping
respacket=swtl.get_grouped(lookup_radii_multiplier=0.9, ht_ratio=3.0)
grouped_annot_bubble = respacket[2]
maskviz = respacket[4]
maskcomb  = respacket[5]

# Saving the results
_=cv2.imwrite('img1_processed.jpg', swtl.swtlabelled_pruned13C)
imgshowN([maskcomb, grouped_annot_bubble], savepath='grouped_img1.jpg')

---

---

Third Image

Notice that the text_mode here is 'db_lf', which stands for Dark Background Light Foreground i.e the text in this image is going to be in lighter color than the background

from swtloc import SWTLocalizer
from swtloc.utils import imgshowN, imgshow

swtl = SWTLocalizer()
# Stroke Width Transform
swtl.swttransform(imgpaths=imgpaths[1], text_mode = 'db_lf',
              save_results=True, save_rootpath = 'swtres/',
              minrsw = 3, maxrsw = 20, max_angledev = np.pi/3)
imgshow(swtl.swtlabelled_pruned13C)

# Grouping
respacket=swtl.get_grouped(lookup_radii_multiplier=0.9, ht_ratio=3.0)
grouped_annot_bubble = respacket[2]
maskviz = respacket[4]
maskcomb  = respacket[5]

# Saving the results
_=cv2.imwrite('img1_processed.jpg', swtl.swtlabelled_pruned13C)
imgshowN([maskcomb, grouped_annot_bubble], savepath='grouped_img1.jpg')

---

You will also notice that the grouping done is not so accurate, to get the desired results as the images might vary, try to tune the grouping parameters in swtl.get_grouped() function.