Extracting text OpenCV
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 t_mean, t_mean_2;
Mat grayF;
outImg_gray.convertTo(grayF, CV_32F);
int winSize = 35;
blur(grayF, t_mean, cv::Size(winSize,winSize));
blur(grayF.mul(grayF), t_mean_2, cv::Size(winSize,winSize));
Mat varMat = t_mean_2 - t_mean.mul(t_mean);
varMat.convertTo(varMat, CV_8U);
// threshold the high variance regions
Mat varMatRegions = varMat > 100;
When given an image like this:
Then when I show varMatRegions I get this image:
As you can see it somewhat combines the left block of text with the header of the card, for most cards this method works great but on busier cards it can cause problems.
The reason it is bad for those contours to connect is that it makes the bounding box of the contour nearly take up the entire card.
Can anyone suggest a different way I can find the text to ensure proper detection of text?
200 points to whoever can find the text in the card above the these two.
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)
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.