1. 云栖社区>
  2. 博客>
  3. 正文


禾路 2018-06-22 22:08:00 浏览1000
一、How to make auto-adjustments(brightness and contrast) for image Android Opencv Image Correction

i'm using OpenCV for Android.
I would like to know,how to make image correction(auto adjustments of brightness/contrast) for image(bitmap) in android via OpenCV or that can be done by native ColorMatrixFilter from Android!??

I tried to google,but didn't found good tutorials/examples.
So how can i achieve my goal? Any ideas?



Brightness and contrast is linear operator with parameter alpha and beta

O(x,y) = alpha * I(x,y) + beta

In OpenCV you can do this with convertTo.

The question here is how to calculate alpha and beta automatically ?

Looking at histogram, alpha operates as color range amplifier, beta operates as range shift.

Automatic brightness and contrast optimization calculates alpha and beta so that the output range is 0..255.

input range = max(I) - min(I)
wanted output range = 255;
alpha = output range / input range = 255 / ( max(I) - min(I) )

min(O) = alpha * min(I) beta
beta = -min(I) * alpha

Histogram Wings Cut (clipping)

To maximize the result of it's useful to cut out some color with few pixels.

This is done cutting left right and wings of histogram where color frequency is less than a value (typically 1%). Calculating cumulative distribution from the histogram you can easly find where to cut.

may be sample chart helps to understand:

image description

By the way BrightnessAndContrastAuto could be named normalizeHist because it works on BGR and gray images stretching the histogram to the full range without touching bins balance. If input image has range 0..255 BrightnessAndContrastAuto will do nothing.

Histogram equalization and CLAE works only on gray images and they change grays level balancing. look at the images below:

Normalization vs Equalization

  void BrightnessAndContrastAuto(const cv::Mat &src, cv::Mat &dst, float clipHistPercent)
    CV_Assert(clipHistPercent >= 0);
    CV_Assert((src.type() == CV_8UC1) || (src.type() == CV_8UC3) || (src.type() == CV_8UC4));
    int histSize = 256;
    float alpha, beta;
    double minGray = 0, maxGray = 0;
    //to calculate grayscale histogram
    cv::Mat gray;
    if (src.type() == CV_8UC1) gray = src;
    else if (src.type() == CV_8UC3) cvtColor(src, gray, CV_BGR2GRAY);
    else if (src.type() == CV_8UC4) cvtColor(src, gray, CV_BGRA2GRAY);
    if (clipHistPercent == 0)
        // keep full available range
        cv::minMaxLoc(gray, &minGray, &maxGray);
        cv::Mat hist; //the grayscale histogram
        float range[] = { 0256 };
        const float* histRange = { range };
        bool uniform = true;
        bool accumulate = false;
        calcHist(&gray, 10, cv::Mat (), hist, 1&histSize, &histRange, uniform, accumulate);
        // calculate cumulative distribution from the histogram
        std::vector<float> accumulator(histSize);
        for (int i = 1; i < histSize; i++)
            accumulator[i] = accumulator[i - 1+<float>(i);
        // locate points that cuts at required value
        float max = accumulator.back();
        clipHistPercent *= (max / 100.0); //make percent as absolute
        clipHistPercent /= 2.0// left and right wings
        // locate left cut
        minGray = 0;
        while (accumulator[minGray] < clipHistPercent)
        // locate right cut
        maxGray = histSize - 1;
        while (accumulator[maxGray] >= (max - clipHistPercent))
    // current range
    float inputRange = maxGray - minGray;
    alpha = (histSize - 1/ inputRange;   // alpha expands current range to histsize range
    beta = -minGray * alpha;             // beta shifts current range so that minGray will go to 0
    // Apply brightness and contrast normalization
    // convertTo operates with saurate_cast
    src.convertTo(dst, -1, alpha, beta);
    // restore alpha channel from source 
    if (dst.type() == CV_8UC4)
        int from_to[] = { 33};
        cv::mixChannels(&src, 4&dst,1, from_to, 1);


二、Template matching behavior - Color

I am evaluating template matching algorithm to differentiate similar and dissimilar objects. What I found is confusing, I had an impression of template matching is a method which compares raw pixel intensity values. Hence when the pixel value varies I expected Template Matching to give a less match percentage.

I have a template and search image having same shape and size differing only in color(Images attached). When I did template matching surprisingly I am getting match percentage greater than 90%.

img = cv2.imread('./images/searchtest.png', cv2.IMREAD_COLOR)template = cv2.imread('./images/template.png', cv2.IMREAD_COLOR)
res = cv2.matchTemplate(img, template, cv2.TM_CCORR_NORMED)
min_val, max_val, min_loc, max_loc = cv2.minMaxLoc(res)print(max_val)

Template Image :
Template Image

Search Image :
Search Image

Can someone give me an insight why it is happening so? I have even tried this in HSV color space, Full BGR image, Full HSV image, Individual channels of B,G,R and Individual channels of H,S,V. In all the cases I am getting a good percentage.

Any help could be really appreciated.

image description

这个问题的核心就是彩色模板匹配。很有趣的问题。回答直接给出了source code,
三、likely position of Feature Matching.

I am using Brute Force Matcher with L2 norm. Referring this link

After the process, I get below image as output

image description

What is the likely position of the object suggested by the feature matching?

I don't understand how to choose the likely position using this image :(



to retrieve the position of your matched object, you need some further steps :

  • filter the matches for outliers
  • extract the 2d point locations from the keypoints
  • apply findHomography() on the matched 2d points to get a transformation matrix between your query and the scene image
  • apply perspectiveTransform on the boundingbox of the query object, to see, where it is located in the scene image.


image description

 //used surf
    #include "stdafx.h"
    #include <iostream>
    #include "opencv2/core/core.hpp"
    #include "opencv2/imgproc/imgproc.hpp"
    #include "opencv2/features2d/features2d.hpp"
    #include "opencv2/highgui/highgui.hpp"
    #include "opencv2/nonfree/features2d.hpp"
    #include "opencv2/calib3d/calib3d.hpp"
    using namespace std;
    using namespace cv;
    int main( int argc, char** argv )
        Mat img_1 ;
        Mat img_2 ;
        Mat img_raw_1 = imread("c1.bmp");
        Mat img_raw_2 = imread("c3.bmp");
        //-- Step 1: 使用SURF识别出特征点
        int minHessian = 400;
        SurfFeatureDetector detector( minHessian );
        std::vector<KeyPoint> keypoints_1, keypoints_2;
        detector.detect( img_1, keypoints_1 );
        detector.detect( img_2, keypoints_2 );
        //-- Step 2: 描述SURF特征
        SurfDescriptorExtractor extractor;
        Mat descriptors_1, descriptors_2;
        extractor.compute( img_1, keypoints_1, descriptors_1 );
        extractor.compute( img_2, keypoints_2, descriptors_2 );
        //-- Step 3: 匹配
        FlannBasedMatcher matcher;//BFMatcher为强制匹配
        std::vector< DMatch > matches;
        matcher.match( descriptors_1, descriptors_2, matches );
        double max_dist = 0; double min_dist = 100;
        for( int i = 0; i < descriptors_1.rows; i++ )
            double dist = matches[i].distance;
            if( dist < min_dist ) min_dist = dist;
            if( dist > max_dist ) max_dist = dist;
        std::vector< DMatch > good_matches;
        for( int i = 0; i < descriptors_1.rows; i++ )
            if( matches[i].distance <= 3*min_dist )//这里的阈值选择了3倍的min_dist
                good_matches.push_back( matches[i]); 
        //-- Localize the object from img_1 in img_2
        std::vector<Point2f> obj;
        std::vector<Point2f> scene;
        for( int i = 0; i < (int)good_matches.size(); i++ )
            scene.push_back( keypoints_1[ good_matches[i].queryIdx ].pt );
            obj.push_back( keypoints_2[ good_matches[i].trainIdx ].pt );
        Mat H = findHomography( obj, scene, CV_RANSAC );
        Mat result;
        Mat half(result,cv::Rect(0,0,img_2.cols,img_2.rows));
        return 0;
四、Run OpenCV to MFC
 I have reproduced this sample, in a MFC app.

The cv::Mat is a CDOcument variable member:

// Attributes
std::vector<CBlob> m_blobs;
cv::Mat m_Mat;

and I draw rectangles over image, with an method (called at some time intervals, according FPS rate):

DrawBlobInfoOnImage(m_blobs, m_Mat);

Here is the code of this method:

void CMyDoc::DrawBlobInfoOnImage(std::vector<CBlob>& blobs, cv::Mat& Mat)
for (unsigned int i = 0;i < blobs.size();++i)
    if (blobs[i].m_bStillBeingTracked)
        cv::rectangle(Mat, blobs[i].m_rectCurrentBounding, SCALAR_RED, 2);
        double dFontScale = blobs[i].m_dCurrentDiagonalSize / 60.0;
        int nFontThickness = (int)roundf(dFontScale * 1.0);
        cv::putText(Mat, (LPCTSTR)IntToString(i), blobs[i].m_vecPointCenterPositions.back(), CV_FONT_HERSHEY_SIMPLEX, dFontScale, SCALAR_GREEN, nFontThickness);

but the result of this method is something like that:

image description

My question is: how can I draw only the last blobs result over my image ?

I have tried to clean up m_Mat, and to enable to draw only blobs.size() - 1 blob over image, none of this worked ...








五、How to find the thickness of the red color sealent in the image ????


I want to find the thickness of the red colored sealent in the image.

First I'm extracting the sealent portion by using findcontours by having minimum and maximum conotur area.And then check the Area,length and thickness of the sealent,i can find the area as well as length but i m not able to find the thickness of the sealent portion.

Please help me guys........below is the example image.


do a distance transform and a skeleton on the binary image.




目前方向:图像拼接融合、图像识别 联系方式


+ 关注