sagoma/lines.cpp

#include<cmath>
#include<algorithm>
#include<unistd.h>
#include<fstream>
#include<iomanip>

#ifdef HAS_OPENCV3
#include <opencv2/core.hpp> //Any OPENCV3 code
#else
#include <opencv2/core/core.hpp> //Any Opencv2 code
#endif
#include"lines.h"
#include "geometry.h"
#include "cvutils.h"
using namespace cv;

#define likely(x) __builtin_expect(!!(x), 1)
#define unlikely(x) __builtin_expect(!!(x), 0)

// all those colors are still visible on b/w (the first component is not so low)
auto WHITE = cv::Scalar(255,255,255);
auto BLUE = cv::Scalar(200,50,50);
auto MAGENTA = cv::Scalar(110,10,200);
auto BROWN = cv::Scalar(90,100,60);
auto BLACK = cv::Scalar(0,0,0);
auto YELLOW = cv::Scalar(20,200,200);

void usage(char* program, std::ostream &buf) {
	buf << "Usage: " << program << "[options] IMAGE-INPUT" << std::endl;
	buf << "Identify the image as a book in perspective, get deperspectivized content" << std::endl;
	buf << std::endl;
	buf << "Options:" << std::endl;
	buf << "  -p     PROFILE     Save the transformation matrix in PROFILE" << std::endl;
	buf << "  -l     LEFTPAGE    Save the left page in LEFTPAGE" << std::endl;
	buf << "  -r     RIGHTPAGE   Save the right page in RIGHTPAGE" << std::endl;
	buf << "  -h     HELP        Show this help and exit" << std::endl;
}

class Options
{
	public:
		char *left;
		char *right;
		char *profile;
		char *input;
		Options();
		Options(int argc, char*argv[]);
		bool parse(int argc, char*argv[]);
		~Options();
};
Options::Options()
{
	left = right = profile = input = NULL;
}
Options::~Options()
{
	if(left == NULL) {
		free(left);
	}
	if(right == NULL) {
		free(right);
	}
	if(profile == NULL) {
		free(profile);
	}
	//input doesn't need to be free-d: it points to argv[argc-1]
}
Options::Options(int argc, char *argv[])
{
	left = right = profile = input = NULL;
	parse(argc, argv);
}

/* Options::parse  parse arguments, return true if help is requested */
bool Options::parse(int argc, char *argv[])
{
	int c;
	while((c=getopt(argc, argv, "p:l:r:h")) != -1)
	{
		switch(c)
		{
			case 'l':
				left = strdup(optarg);
				break;
			case 'r':
				right = strdup(optarg);
				break;
			case 'p':
				profile = strdup(optarg);
				break;
			case 'h':
				return true;
				break;
			default:
				throw std::runtime_error("Parsing error: invalid argument");
		}
	}
	if(optind >= argc) {
		std::cerr << "Error: " << argv[0] << " needs an argument" << std::endl;
		throw std::runtime_error("Parsing error: argument needed");
	}
	input = argv[optind++];
	if(optind < argc) {
		std::cerr << "Error: too many arguments supplied" << std::endl;
		throw std::runtime_error("Parsing error: too many arguments");
	}
	return false;
}

class BookShape {
	private:
		cv::Mat trasf[2];
		void precompute();
		cv::Point2f trapezoids[2][4]; //transformation matrix
	public:
		BookShape(cv::Point tl, cv::Point tm, cv::Point tf,
				             cv::Point bl, cv::Point bm, cv::Point br);
		double xsize();
		double ysize();
		cv::Mat* getTrasfs();
};
//topleft topmiddle topright
//botleft botmiddle botright
BookShape::BookShape(cv::Point tl, cv::Point tm, cv::Point tr, cv::Point bl, cv::Point bm, cv::Point br)
{
	trapezoids[0][0] = tl;
	trapezoids[0][1] = tm;
	trapezoids[0][2] = bm;
	trapezoids[0][3] = bl;
	trapezoids[1][0] = tm;
	trapezoids[1][1] = tr;
	trapezoids[1][2] = br;
	trapezoids[1][3] = bm;
	precompute();
}

void BookShape::precompute(void) {
	cv::Point2f outsizes[4] = {cv::Point2f(0, 0), cv::Point(xsize(), 0),
		cv::Point(xsize(), ysize()), cv::Point(0, ysize())};
	for(int i=0; i < 2; i++) {
		trasf[i] = cv::getPerspectiveTransform(trapezoids[i], outsizes);
		assert(9==trasf[i].total()); // 3x3
	}
}
cv::Mat* BookShape::getTrasfs()
{
	return trasf;
}
double BookShape::xsize()
{
	return dist(trapezoids[0][0], trapezoids[0][1]);
}
double BookShape::ysize()
{
	return dist(trapezoids[0][1], trapezoids[0][2]);
}


BookShape get_book_shape(cv::Mat img)
{
	//dotwidth is just a simple size so that lines and dots are visible on big images
	//but not huge on small images
#ifdef _DEBUG
	unsigned short dotwidth = img.cols >> 6;  // divide by 64, so efficient
#endif

	std::vector< std::vector<cv::Point> > contours;
	std::vector<cv::Vec4i> hierarchy; //this is not really useful
	cv::findContours(img,contours,hierarchy,CV_RETR_EXTERNAL,CV_CHAIN_APPROX_SIMPLE);


	if( 0==contours.size() ) {
		std::cerr << "No contours found" << std::endl;
		throw "Error getting contours";
	}
	auto contour = contours[0];
	if( 1!=contours.size() )
	{
		// choosing the biggest contour in the image
		// you can test it with files/2contours.png
		for(auto cont: contours) {
			if(cont.size() > contour.size())
				contour = cont;
		}
	}

	//hull: points forming an external convex border
	std::vector<cv::Point> hull;
	cv::convexHull(cv::Mat(contour),hull,false);

	//because of the "butterfly" shape of a book, the most distant points in the hull
	//are the one between the corner.
	unsigned *maxdistances = max2_distance(hull);
	cv::Point corn_1, corn_2, corn_3, corn_4;
	corn_1 = hull[maxdistances[0]];
	corn_2 = hull[(maxdistances[0]+1)%hull.size()];
	corn_3 = hull[maxdistances[1]];
	corn_4 = hull[(maxdistances[1]+1)%hull.size()];

	std::vector<cv::Point> vertical_points[2];
	cv::Vec<double,3> verticals[2];
	// Between the two corners on the same side, the longest line is the vertical border of the book
	vertical_points[0] = find_longest_line(hull, (maxdistances[0]+1)%hull.size(), maxdistances[1]);
	std::cout << maxdistances[1] << std::endl;
	std::cout << maxdistances[1]+1 << std::endl;
	std::cout << (maxdistances[1]+1)%hull.size() << std::endl;
	vertical_points[1] = find_longest_line(hull, (maxdistances[1]+1)%hull.size(), maxdistances[0]);
	free(maxdistances);
#ifdef _DEBUG
	img = cv::imread("files/masckera.png", CV_LOAD_IMAGE_COLOR);
	cv::circle(img, vertical_points[0][0], img.cols>>7, cv::Scalar(200, 0, 0), -1);
	cv::circle(img, vertical_points[0][vertical_points[0].size()-1], img.cols>>7, cv::Scalar(200, 0, 0), -1);
	cv::circle(img, vertical_points[1][0], img.cols>>7, cv::Scalar(200, 200, 0), -1);
	cv::circle(img, vertical_points[1][vertical_points[1].size()-1], img.cols>>7, cv::Scalar(200, 0, 0), -1);
	cv::circle(img, corn_1, img.cols>>7, cv::Scalar(200, 0, 200), -1);
	cv::namedWindow("aaa", CV_GUI_NORMAL);
	cv::imshow("aaa", img);
	cv::waitKey(0);
#endif
	verticals[0] = get_line(vertical_points[0][0], vertical_points[0][vertical_points[0].size()-1]);
	std::cout << verticals[0] << std::endl;
	std::cout << line_value(verticals[0], corn_1) << std::endl;
	verticals[0][2] -= line_value(verticals[0], corn_1);
	std::cout << line_value(verticals[0], corn_1) << "=0?" << std::endl;
	verticals[1] = get_line(vertical_points[1][0], vertical_points[1][vertical_points[1].size()-1]);

	// theta is the angle of the line connecting point 1 and 2; it will be the
	// rotation of our new coordinate system
	auto cs = CoordinateSystem(corn_1, corn_2);
	assert(cs.map(corn_1).x == 0);
	assert(cs.map(corn_1).y == 0);
	assert(cs.map(corn_2).x > 0);
	assert(cs.map(corn_2).y == 0);
	cv::Vec<double,3> diag1, diag2;
	diag1 = get_line(cs.map(corn_1), cs.map(corn_3));
	diag2 = get_line(cs.map(corn_4), cs.map(corn_2));
	std::cout << "mapped diag1: " << diag1 << std::endl;
	std::cout << "mapped diag2: " << diag2 << std::endl;

	std::vector<cv::Point> points1, points2;
	for(cv::Point p: contour)
	{ // the point is interesting where it is above both lines or below both lines
		cv::Point mapped = cs.map(p);
		if(is_above_line(diag1, mapped)) {
		   if(is_above_line(diag2, mapped)) {
			points1.push_back(p);
		   }
		} else {
		   if(!is_above_line(diag2, mapped)) {
			points2.push_back(p);
		   }
		}
	}

	cv::Point middle1 = further_point_from_line(get_line(corn_1, corn_2), points2);
	cv::Point middle2 = further_point_from_line(get_line(corn_3, corn_4), points1);

	BookShape bs = BookShape(
			vertical_points[1][vertical_points[1].size()-1], middle1, vertical_points[0][0],
			vertical_points[1][0], middle2, vertical_points[0][vertical_points[0].size()-1]
			);
	return bs;
}

int main(int argc, char *argv[])
{
	Options args;
	try{
		if(args.parse(argc, argv)) {
			usage(argv[0], std::cout);
			return EXIT_SUCCESS;
		}
	} catch(std::runtime_error) {
		usage(argv[0], std::cerr);
		return EXIT_FAILURE;
	}

	cv::Mat img;
	img=cv::imread(args.input,CV_LOAD_IMAGE_GRAYSCALE);
	if( img.empty() ) {
		std::cerr << "Error opening image, aborting" << std::endl;
		return EXIT_FAILURE;
	}
	BookShape bs = get_book_shape(img);

	if(args.profile) {
		// format: each line is a tab-delimited 3x3 transformation matrix (as get by getPerspectiveTransform)
		//    first three elements is the first line, etc
		std::ofstream profilebuf;
		profilebuf.open(args.profile);
		for(int t=0; t<2; t++) {
			for(int i=0; i<3; i++) {
				double* row= bs.getTrasfs()[t].ptr<double>(i);
				for(int j=0; j < 3; j++) {
					profilebuf << std::fixed << std::setprecision(16);
					profilebuf << row[j] << "\t"; //yes, there's a trailing tab
				}
			}
			profilebuf << std::endl;
		}
		profilebuf.close();
	}

	//TODO: distinguish left and right
	std::vector<int> params;
	if(args.left || args.right) {
		//we are rereading in full color to get colored output
		img = cv::imread(args.input,CV_LOAD_IMAGE_COLOR);
		cv::Mat rect[2]; //final pages, transformed in a nice rectangle

		params.push_back(CV_IMWRITE_PNG_COMPRESSION);
		params.push_back(9);
		if(args.left) {
			cv::warpPerspective(img, rect[0], bs.getTrasfs()[0], cv::Size(bs.xsize(), bs.ysize()));
			cv::imwrite(args.left, rect[0], params);
		}
		if(args.right) {
			cv::warpPerspective(img, rect[1], bs.getTrasfs()[1], cv::Size(bs.xsize(), bs.ysize()));
			cv::imwrite(args.right, rect[1], params);
		}
	}

	return EXIT_SUCCESS;
}

// vim: set noet ts=4 sw=4: