Introduction to C & C++

Lecture 10 – Image Processing with OpenCV

JJCAO

OpenCV (Open Source Computer Vision) is a library of programming functions for real time computer vision.

Supported Development Environment

• Linux, Windows, Android• GCC, Eclipse, Visual Studio,

Install OpenCV

1. Basic Info: http://opencv.willowgarage.com/wiki/InstallGuide

2. Necessary steps for Qt Creator (Similar with VS): http://www.laganiere.name/opencvCookbook/chap1.shtml

3. Installation in Windows (too much of details): http://opencv.itseez.com/doc/tutorials/introduction/windows_install/windows_install.html#windows-installation

How to build applications with OpenCV inside the Microsoft Visual Studio

http://opencv.itseez.com/doc/tutorials/introduction/windows_visual_studio_Opencv/windows_visual_studio_Opencv.html#windows-visual-studio-how-to

P11 of OpenCV.2.Computer.Vision.Application.Programming.Cookbook_2011

Load & Display an Image

1. Load an image (using imread)

2. Create a named OpenCV window (using namedWindow)

3. Display an image in an OpenCV window (using imshow)

1. Load an image (using imread)cv::Mat image;image = cv::imread(argv[1], CV_LOAD_IMAGE_COLOR); // Read the file

if( img.empty() ) // Check for invalid input{std::cout << "Could not open or find the image" << std::endl ;return -1;}

Image formats supported: Bmp, jpg, png, tif, ppm, …

CV_LOAD_IMAGE_UNCHANGED (<0) loads the image as is (including the alpha channel if present)

CV_LOAD_IMAGE_GRAYSCALE (0) loads the image as an intensity one

CV_LOAD_IMAGE_COLOR (>0) loads the image in the RGB format

2. Create a named OpenCV window (using namedWindow)

cv::namedWindow( "Display window", CV_WINDOW_NORMAL|CV_WINDOW_FREERATIO );// Create a window for display.

CV_WINDOW_AUTOSIZE is the only supported one if you do not use the Qt backend. In this case the window size will take up the size of the image it shows. No resize permitted!CV_WINDOW_NORMALon Qt you may use this to allow window resize. The image will resize itself according to the current window size. By using the | operator you also need to specify if you would like the image to keep its aspect ratio (CV_WINDOW_KEEPRATIO) or not (CV_WINDOW_FREERATIO).

3. Display an image in an OpenCV window (using imshow)

cv::imshow( "Display window", image ); // Show our image inside it.

cv::waitKey(0); // Wait for a keystroke in the window

• Because we want our window to be displayed until the user presses a key (otherwise the program would end far too quickly), we use the waitKey function whose only parameter is just how long should it wait for a user input (measured in milliseconds).

• Zero means to wait forever.

Necessary Head Files#include <opencv2/core/core.hpp>#include <opencv2/highgui/highgui.hpp>using namespace cv;using namespace std;

You’ll almost always end up using the:• core section, as here are defined the basic building blocks of the

library• highgui module, as this contains the functions for input, output

& GUI operations

Necessary libraries

#ifdef _DEBUG#pragma comment(lib, "opencv_core230d.lib“ ) #pragma comment(lib, "opencv_highgui230d.lib“ ) #else#pragma comment(lib, "opencv_core230.lib“ ) #pragma comment(lib, "opencv_highgui230.lib“ ) #endif

Congratulation!

From C to C++• OpenCV has been around ever since 2001. In those days

the library was built around a C interface.– user is responsible for taking care of memory allocation and de-

allocation– Lots of old tutorials written in C

• Once your code start to grow larger & larger, more & more a struggle to handle this rather than focusing on solving your goal

• Finally C++– automatic memory management (more or less)– less to write, to achieve more

class CV_EXPORTS Mat { public: // ... a lot of methods ... ... /*! includes several bit-fields: - the magic signature - continuity flag- depth - number of channels */

int flags; int dims; //! the array dimensionality, >= 2

int rows, cols; //! the number of rows and columns or (-1, -1) when the array has more than 2 dimensions

int* refcount; //! pointer to the reference counter; when array points to user-allocated data, the pointer is NULL// other members ...

uchar* data; //! pointer to the data

};

Head

Create a cv::Mat

Mat A, C; // creates just the header parts

A = imread(argv[1], CV_LOAD_IMAGE_COLOR); // the method will allocate matrix (deep copy)

Mat B(A); // Use the copy constructor, without copying the data (shadow copy)C = A; // Assignment operator, shadow copy

Mat roi(A, Rect(10,10,100,100)); // select a ROI roi = Scalar(0,255,0); // fill the ROI with (0,255,0) (which is green in RGB space); the original A will be modified; see next page.

cv::Scalar

template <typename _Tp> class Scalar_ : public Vec<_Tp, 4> { ... };

typedef Scalar_<double> Scalar;

Being derived from Vec<_Tp, 4> , Scalar_ and Scalar can be used just as typical 4-element vectors.

Deep Copy

Mat F = A.clone(); Mat G; A.copyTo(G);

Now modifying F or G will not affect the matrix pointed by the Mat header.

What you need to remember1. Output image allocation for OpenCV functions is automatic

(unless specified otherwise).– Example (next page)

2. No need to think about memory freeing with OpenCVs C++ interface.

3. The assignment operator and the copy constructor (ctor)copies only the header.

4. Use the clone() or the copyTo() function to copy the underlying matrix of an image.

Output image allocation for OpenCV functions is automatic

• instead of writing:Mat color; ... Mat gray(color.rows, color.cols, color.depth()); cvtColor(color, gray, CV_BGR2GRAY);

• you can simply write:Mat color; ... Mat gray; cvtColor(color, gray, CV_BGR2GRAY);

How to scan images, lookup table & time measurement

• How to go through each and every pixel of an image?• How is OpenCV matrix values stored?• How to measure the performance of our algorithm?• What are lookup tables and why use them?

1. Basic Mat info2. Storing methods3. Data type conversion4. Accessing Pixel Values

a simple color reduction method

how_to_scan_images imageName.jpg divideWith [G]

if( argc == 4 && !strcmp(argv[3],"G") ) I = imread(argv[1], CV_LOAD_IMAGE_GRAYSCALE); else I = imread(argv[1], CV_LOAD_IMAGE_COLOR);

uchar

CV_8U

CV_8U3

Storing methods• How to store pixel values?– Color space

• Gray-level (Black-and-White)– unsigned 8-bit values– 0: black, 255: white

• RGB– A triplet of unsigned 8-bit values– [0, 0, 0]: black, [255, 0, 0]: red, [255, 255, 255]: white

• HSV, HLS, …, CIE– Data type

bool Mat::isContinuous()

• Matrices created with Mat::create() are always continuous:

• may no longer continuous– extract a part of the matrix using Mat::col(), Mat::

diag() , and so on, – or constructed a matrix header for externally

allocated data.

row1 row2 … row n

Storing methods• How to store pixel values?– Color space

• Gray-level (Black-and-White)– unsigned 8-bit values– 0: black, 255: white

• RGB– A triplet of unsigned 8-bit values– [0, 0, 0]: black, [255, 0, 0]: red, [255, 255, 255]: white

• HSV, HLS, …, CIE– Data type

CV_8U - 8-bit unsigned integers ( 0..255 )CV_8S - 8-bit signed integers ( -128..127 )CV_16U - 16-bit unsigned integers ( 0..65535 )CV_16S - 16-bit signed integers ( -32768..32767 )CV_32S - 32-bit signed integers ( -2147483648..2147483647 )CV_32F - 32-bit floating-point numbers ( -FLT_MAX..FLT_MAX, INF, NAN )CV_64F - 64-bit floating-point numbers ( -DBL_MAX..DBL_MAX, INF, NAN )

Inquire Mat Info

1. int Mat::depth() const2. int Mat::channels() const3. int Mat::type() // mixture of depth & channels

– #define CV_8UC1 CV_MAKETYPE(CV_8U,1)– #define CV_8UC2 CV_MAKETYPE(CV_8U,2)– #define CV_8UC3 CV_MAKETYPE(CV_8U,3)– #define CV_8UC4 CV_MAKETYPE(CV_8U,4)– #define CV_8UC(n) CV_MAKETYPE(CV_8U,(n))– …

4. size_t Mat::elemSize() // matrix element (pixel) size in bytes

– if the matrix type is CV_16SC3 , the method returns3*sizeof(short) or 6.

5. size_t Mat::elemSize1() // element size of a channel in bytes

– if the matrix type is CV_16SC3 , the method returns sizeof(short) or 2.6. M.step[]7. size_t Mat::step1() // a matrix step divided by Mat::elemSize1()

Accessing Pixel Values

1. The efficient way: c style access p = I.ptr<uchar>(i); p[j]

2. The iterator (safe) method3. On-the-fly address calculation with reference returning

I.at<uchar>(i,j)

• Performance Differencequite large (2560 X 1600) image

Performance Difference

Debug

Release

More efficient Algorithm

1. divide and multiplication operations are bloody expensive for a system.

2. cheaper operations such as a few subtractions, addition or in best case a simple assignment

3. limited number of input values, 256 to be exact in this problem

Lookup table int divideWith; // convert our input string to number - C++ style stringstream s; s << argv[2]; s >> divideWith; if (!s) { cout << "Invalid number entered for dividing. " << endl; return -1; } uchar table[256]; for (int i = 0; i < 256; ++i) table[i] = divideWith* (i/divideWith);

Mat& ScanImageAndReduceC(Mat& I, const uchar* const table)

{ // accept only char type matrices CV_Assert(I.depth() != sizeof(uchar));

int channels = I.channels();

int nRows = I.rows; int nCols = I.cols * channels;

if (I.isContinuous()) { nCols *= nRows; nRows = 1; }

int i,j; uchar* p; for( i = 0; i < nRows; ++i) { p = I.ptr<uchar>(i); for ( j = 0; j < nCols; ++j) { p[j] = table[p[j]]; } } return I; }

Basic Mat Info

1. bool Mat::empty()2. size_t Mat::total()3. int Mat::rows, Mat::cols4. Size Mat::size()

Mat& ScanImageAndReduceIterator(Mat& I, const uchar* const table)

{ // accept only char type matrices CV_Assert(I.depth() != sizeof(uchar)); const int channels = I.channels(); switch(channels) {

case 1: { MatIterator_<uchar> it, end; for( it = I.begin<uchar>(), end = I.end<uchar>(); it != end; ++it) *it = table[*it]; break; }

case 3: { MatIterator_<Vec3b> it, end; for( it = I.begin<Vec3b>(), end = I.end<Vec3b>(); it != end; ++it) { (*it)[0] = table[(*it)[0]]; (*it)[1] = table[(*it)[1]]; (*it)[2] = table[(*it)[2]]; } } } return I; }

Mat& ScanImageAndReduceRandomAccess(Mat& I, const uchar* const table)

{ // accept only char type matrices CV_Assert(I.depth() != sizeof(uchar));

const int channels = I.channels(); switch(channels) {

case 1: { for( int i = 0; i < I.rows; ++i) for( int j = 0; j < I.cols; ++j ){ I.at<uchar>(i,j) = table[I.at<uchar>(i,j)]; } break; }

case 3: { Mat_<Vec3b> _I = I; for( int i = 0; i < I.rows; ++i) for( int j = 0; j < I.cols; ++j ) { _I(i,j)[0] = table[_I(i,j)[0]]; _I(i,j)[1] = table[_I(i,j)[1]]; _I(i,j)[2] = table[_I(i,j)[2]]; } I = _I; break; } } return I;}

The Core Function: cv::LUT()

LUT: replace all of given image values to some other values Mat lookUpTable(1, 256, CV_8U); uchar* p = lookUpTable.data; for( int i = 0; i < 256; ++i) p[i] = table[i]; for (int i = 0; i < times; ++i) cv::LUT(I, lookUpTable, J);

Conclusion1. If possible, use the already made functions of OpenCV (instead

reinventing these).

2. The fastest method turns out to be the LUT function. This is because the OpenCV library is multi-thread enabled via Intel Threaded Building Blocks.

3. However, if you need to write a simple image scan prefer the pointer method. The iterator is a safer bet, however quite slower.

4. Using the on-the-fly reference access method for full image scan is the most costly in debug mode. In the release mode it may beat the iterator approach or not, however it surely sacrifices for this the safety trait of iterators.

Type Conversion

#include <opencv2/imgproc/imgproc.hpp>

cvtColor(I, J, CV_RGB2GRAY);for( int i = 0; i < 0.5*I.rows; ++i)for( int j = 0; j < 0.5*I.cols; ++j ) {

//J.at<uchar>(i,j) = 0;J.at<float>(i,j) = 0;

}

Mat::convertTo()

void Mat::convertTo(OutputArray m, int rtype, double alpha=1, double beta=0 )

• Converts an array to another datatype with optional scaling.• Parameters:

– m – Destination matrix. If it does not have a proper size or type before the operation, it is reallocated.

– rtype – Desired destination matrix type or, rather, the depth since the number of channels are the same as the source has. If rtype is negative, the destination matrix will have the same type as the source.

– alpha – Optional scale factor.– beta – Optional delta added to the scaled values.

• The method converts source pixel values to the target datatype. saturate_cast<> is applied at the end to avoid possible overflows:

One More Example: gradientinfile='data\HappyFish.jpg';im = imread(infile);info = imfinfo(infile);if ~strcmp(info.ColorType,'grayscale') im = double(rgb2gray(im)) ;end[gx,gy] = gradient(im); figure;colormap(gray);imagesc(im);hold on;[x,y] = meshgrid(1:n,1:m);quiver(x, y, gx,gy);

See the cvMatlab and cvMatlabTest example

void jj::gradient(cv::InputArray _src, cv::OutputArray _dst, int xorder){cv::Mat src=_src.getMat();_dst.create( src.size(), CV_MAKETYPE(src.depth(), src.channels()) ); cv::Mat dst = _dst.getMat();…}

See the cvMatlab and cvMatlabTest example

• Difference in x direction// single channelMat src, dest; // same size, diff depthfor( int i = 0; i < src.rows; ++i)

for( int j = 1; j < src.cols-1; ++j ) {dest.at<uchar>(i,j) = (src.at<uchar>(i,j+1) - src.at<uchar>(i,j-1) ) /2.0;

}// 3 channelsdst.at<cv::Vec3b>(i,j)[0] = (src.at<cv::Vec3b>(i,j+1)[0] - src.at<cv::Vec3b>(i,j-1)[0] ) /2.0;

dst.col(j)=(src.col(j+1)-src.col(j-1))/2.0;

Create a Mat object 1• Mat()Mat M(2,2, CV_8UC3, Scalar(0,0,255)); cout << "M = " << endl << " " << M << endl << endl;

• Create a matrix with more than two dimensionsint sz[3] = {2,2,2}; Mat L(3,sz, CV_8UC(1), Scalar::all(0)); // Specify its dimension, then pass a pointer containing the size for each dimension and the rest remains the same.

• Create a header for an already existing IplImage pointerIplImage* img = cvLoadImage("greatwave.png", 1); Mat mtx(img); // convert IplImage* -> Mat

Create a Mat object 2• Create() functionM.create(4,4, CV_8UC(2)); cout << "M = "<< endl << " " << M << endl << endl;// You cannot initialize the matrix values with this construction. It will only reallocate its matrix data memory if the new size will not fit into the old one.

• MATLAB style initializer: zeros(), ones(), :eyes()Mat E = Mat::eye(4, 4, CV_64F); Mat O = Mat::ones(2, 2, CV_32F); Mat Z = Mat::zeros(3,3, CV_8UC1);

Create a Mat object 3

• For small matricesMat C = (Mat_<double>(3,3) << 0, -1, 0, -1, 5, -1, 0, -1, 0);

• Create a new header for an existing Mat object and clone() or copyTo() it

Mat RowClone = C.row(1).clone();

Create a Mat object 4• Create a random matrix with randu()Mat R = Mat(3, 2, CV_8UC3); randu(R, Scalar::all(0), Scalar::all(255));cout << "R (default) = " << endl << R << endl << endl;

cout << "R (python) = " << endl << format(R,"python") << endl << endl;

cout << "R (csv) = " << endl << format(R,"csv" ) << endl << endl;…

Resources• http://opencv.willowgarage.com/wiki/• C++ Cheatsheet.pdf• Online reference manual: 2.3 documentation is here

20112008

References

• OpenCV 2 Computer Vision Application Programming Cookbook, 2011.