twain3.0/huagao/ImageProcess/ImageApplyAutoCrop.cpp

#include "ImageApplyAutoCrop.h"
#include "ImageProcess_Public.h"

CImageApplyAutoCrop::CImageApplyAutoCrop()
	: m_isCrop(false)
	, m_isDesaskew(false)
	, m_isFillBlank(false)
	, m_isConvexHull(true)
	, m_isFillColor(false)
	, m_threshold(40)
	, m_noise(8)
	, m_indent(5)
	, m_normalCrop(false)
{
}

CImageApplyAutoCrop::CImageApplyAutoCrop(bool isCrop, bool isDesaskew, bool isFillBlank, const cv::Size& fixedSize, bool isConvex, bool isFillColor,
	double threshold, int noise, int indent, bool normalCrop)
	: m_isCrop(isCrop)
	, m_isDesaskew(isDesaskew)
	, m_isFillBlank(isFillBlank)
	, m_isConvexHull(isConvex)
	, m_isFillColor(isFillColor)
	, m_threshold(threshold)
	, m_noise(noise)
	, m_indent(indent)
	, m_fixedSize(fixedSize)
	, m_normalCrop(normalCrop)
{
}

CImageApplyAutoCrop::~CImageApplyAutoCrop()
{
}

cv::Mat concatenateMatrix(const cv::Mat& first, const cv::Mat& second)
{
	cv::Mat mul1 = cv::Mat::eye(3, 3, CV_64F);
	cv::Mat mul2 = cv::Mat::eye(3, 3, CV_64F);
	cv::Mat mul_r;
	first.convertTo(mul_r, CV_64F);
	mul_r.row(0).copyTo(mul1.row(0));
	mul_r.row(1).copyTo(mul1.row(1));

	second.convertTo(mul_r, CV_64F);
	mul_r.row(0).copyTo(mul2.row(0));
	mul_r.row(1).copyTo(mul2.row(1));

	mul1 = mul2 * mul1;
	mul_r = first.clone();
	mul1.row(0).copyTo(mul_r.row(0));
	mul1.row(1).copyTo(mul_r.row(1));
	return mul_r;
}

std::vector<cv::Mat> comMat()
{
	std::vector<cv::Mat> mats;
	cv::Point2f srcTri[3];
	srcTri[0] = cv::Point2f(1, 1);
	srcTri[1] = cv::Point2f(1, 0);
	srcTri[2] = cv::Point2f(0, 1);
	const float fact = 0.33f;
	float pos[] = { 0, 2 * fact, fact };	
	cv::Point2f dstTri[3];
	for (int i = 0; i < 3; i++)
	{
		dstTri[0] = cv::Point2f(1, 1 + pos[i]);
		dstTri[1] = cv::Point2f(1, pos[i]);
		dstTri[2] = cv::Point2f(0, 1 + pos[i]);

		mats.push_back(cv::getAffineTransform(srcTri, dstTri));
	}
	return mats;
}

void brightSharp(cv::Mat& src)
{
	const float a = -0.49f;
	const float b = 3.0f;
	//float kernel_data[] = {
	//  a, 0, 0, 0, a,
	//  0, 0, a, 0, 0,
	//  0, a, b, a, 0,
	//  0, 0, a, 0, 0,
	//  a, 0, 0, 0, a };

	float kernel_data[] = {
	  0, a, 0,
	  a, b, a,
	  0, a, 0
	};
	cv::Mat kernel(3, 3, CV_32FC1, kernel_data);
	cv::filter2D(src, src, src.depth(), kernel);
}

void CImageApplyAutoCrop::apply(cv::Mat& pDib, int side)
{
	(void)side;
	if (pDib.empty()) return;

	if (m_normalCrop)
	{
		cv::Rect roi = cv::Rect((pDib.cols - m_fixedSize.width) / 2, side == 0 ? 75 : 145, m_fixedSize.width, m_fixedSize.height) & cv::Rect(0, 0, pDib.cols, pDib.rows);
		pDib = pDib(roi).clone();
		m_rect = cv::RotatedRect(cv::Point2f(roi.x + roi.width / 2, roi.y + roi.height / 2), cv::Size2f(roi.width, roi.height), 0.0f);
		return;
	}

	if (!m_isCrop && !m_isDesaskew && !m_isFillBlank && m_fixedSize.empty()) return;

	cv::Mat src = pDib;
	cv::Mat thre;
	cv::Mat dst;
	hg::threshold_Mat(src, thre, m_threshold);

	if (m_noise > 0)
	{
		cv::Mat element = getStructuringElement(cv::MORPH_RECT, cv::Size(m_noise, 1));
		cv::morphologyEx(thre, thre, cv::MORPH_OPEN, element, cv::Point(-1, -1), 1, cv::BORDER_CONSTANT, cv::Scalar::all(0));
	}

	if (m_indent > 0)
	{
		cv::Mat element = getStructuringElement(cv::MORPH_ELLIPSE, cv::Size(m_indent, m_indent));
		cv::morphologyEx(thre, thre, cv::MORPH_ERODE, element, cv::Point(-1, -1), 1, cv::BORDER_CONSTANT, cv::Scalar::all(0));
	}

	std::vector<cv::Vec4i> hierarchy;
	std::vector<std::vector<cv::Point>> contours;

	hg::findContours(thre, contours, hierarchy, cv::RETR_EXTERNAL);
	m_maxContour = hg::getMaxContour(contours, hierarchy);

	if (m_maxContour.size() == 0)
	{
		thre.release();
		//
		if (!m_isCrop)
			pDib = pDib(cv::Rect((pDib.cols - m_fixedSize.width) / 2, (pDib.rows - m_fixedSize.height) / 2, m_fixedSize.width, m_fixedSize.height) & cv::Rect(0, 0, pDib.cols, pDib.rows)).clone();
#ifdef LOG
		FileTools::write_log("imgprc.txt", "exit CImageApplyAutoCrop apply");
#endif // LOG
		return;
	}

	thre.release();
	dst.release();

	cv::RotatedRect rect = hg::getBoundingRect(m_maxContour);
	m_rect = rect;
	cv::Rect boudingRect = cv::boundingRect(m_maxContour);
	boudingRect.x -= 1;
	boudingRect.y -= 1;
	boudingRect.width += 2;
	boudingRect.height += 2;

	if (m_isDesaskew && rect.angle != 0)
	{
		cv::Point2f srcTri[4], srcTri_temp[3], dstTri[3];
		rect.points(srcTri);

		dstTri[0] = cv::Point2f(0, rect.size.height - 1);
		dstTri[1] = cv::Point2f(0, 0);
		dstTri[2] = cv::Point2f(rect.size.width - 1, 0);

		srcTri_temp[0] = dstTri[0];
		srcTri_temp[1] = dstTri[1];
		srcTri_temp[2] = dstTri[2];
		cv::Mat warp_mat;
		warp_mat = cv::getAffineTransform(srcTri, dstTri);
		if (src.channels() == 1)
		{
			cv::warpAffine(src, dst, warp_mat, rect.size, cv::INTER_LINEAR);
		}
		else
		{
			cv::Mat bgr[3];
			cv::split(src, bgr);
			auto mats = comMat();
			warp_mat = cv::getAffineTransform(srcTri, dstTri);
			warp_mat = concatenateMatrix(mats[0], warp_mat);

			cv::warpAffine(bgr[0], bgr[0], warp_mat, rect.size, cv::INTER_LINEAR);

			warp_mat = cv::getAffineTransform(srcTri, dstTri);
			warp_mat = concatenateMatrix(mats[1], warp_mat);
			cv::warpAffine(bgr[1], bgr[1], warp_mat, rect.size, cv::INTER_LINEAR);

			warp_mat = cv::getAffineTransform(srcTri, dstTri);
			warp_mat = concatenateMatrix(mats[2], warp_mat);
			cv::warpAffine(bgr[2], bgr[2], warp_mat, rect.size, cv::INTER_LINEAR);

			cv::merge(bgr, 3, dst);
		}

		double* ptr_m = reinterpret_cast<double*>(warp_mat.data);
		double a = ptr_m[0];
		double b = ptr_m[1];
		double c = ptr_m[2];
		double d = ptr_m[3];
		double e = ptr_m[4];
		double f = ptr_m[5];

		for (cv::Point& p : m_maxContour)
		{
			p.x = static_cast<int>(a * p.x + b * p.y + c);
			p.y = static_cast<int>(d * p.x + e * p.y + f);
		}

		for (std::vector<cv::Point>& sub : contours)
			for (cv::Point& p : sub)
			{
				p.x = static_cast<int>(a * p.x + b * p.y + c);
				p.y = static_cast<int>(d * p.x + e * p.y + f);
			}
	}
	else
	{
		auto t_rect = boudingRect & cv::Rect(0, 0, src.cols, src.rows);
		dst = src(t_rect);
		if (dst.channels() == 3)
		{
			cv::Mat bgr[3];
			cv::split(dst, bgr);
			auto mats = comMat();
			for (int i = 0; i < 3; i++)
				cv::warpAffine(bgr[i], bgr[i], mats[i], t_rect.size(), cv::INTER_LINEAR);
			cv::merge(bgr, 3, dst);
		}

		m_maxContour.clear();
		m_maxContour.push_back(cv::Point(0, t_rect.height - 1));
		m_maxContour.push_back(cv::Point(0, 0));
		m_maxContour.push_back(cv::Point(t_rect.width - 1, 0));
		m_maxContour.push_back(cv::Point(t_rect.width - 1, t_rect.height - 1));

		contours.clear();
		contours.push_back(m_maxContour);
	}

	cv::Scalar autoBGColor;
	if (m_isFillBlank)
	{
		if (m_isConvexHull)
		{
			if (m_maxContour.size() == 0)
			{
				thre.release();

				if (!m_isCrop)
					pDib = pDib(cv::Rect((pDib.cols - m_fixedSize.width) / 2, (pDib.rows - m_fixedSize.height) / 2, m_fixedSize.width, m_fixedSize.height) & cv::Rect(0, 0, pDib.cols, pDib.rows)).clone();
				return;
			}
			hg::convexHull(m_maxContour, m_maxContour);
			contours.clear();
			contours.push_back(m_maxContour);
		}

		contours.push_back(std::vector<cv::Point>());
		contours[contours.size() - 1].push_back(cv::Point(-1, dst.rows - 1));
		contours[contours.size() - 1].push_back(cv::Point(-1, -1));
		contours[contours.size() - 1].push_back(cv::Point(dst.cols, -1));
		contours[contours.size() - 1].push_back(cv::Point(dst.cols, dst.rows));

		autoBGColor = m_isFillColor ? getBackGroudColor(pDib, rect.size.area()) : cv::Scalar(255, 255, 255);
		hg::fillPolys(dst, contours, autoBGColor);
	}
	else
	{
		m_maxContour.clear();
		m_maxContour.push_back(cv::Point(-1, dst.rows));
		m_maxContour.push_back(cv::Point(-1, -1));
		m_maxContour.push_back(cv::Point(dst.cols, -1));
		m_maxContour.push_back(cv::Point(dst.cols, dst.rows));
	}

	pDib.release();
	if (/*(m_isCrop && side == 0) || (side == 1 && m_fixedSize.width * m_fixedSize.height == 0)*/ m_isCrop)
		pDib = dst.clone();
	else
	{
		pDib = cv::Mat(m_fixedSize, dst.type(), m_isFillBlank ? autoBGColor : cv::Scalar(0, 0, 0));

		cv::Rect roi;
		roi.x = dst.cols > pDib.cols ? (dst.cols - pDib.cols) / 2 : 0;
		roi.width = cv::min(pDib.cols, dst.cols);
		roi.y = dst.rows > pDib.rows ? (dst.rows - pDib.rows) / 2 : 0;
		roi.height = cv::min(pDib.rows, dst.rows);
		cv::Rect rect((pDib.cols - roi.width) / 2, (pDib.rows - roi.height) / 2, roi.width, roi.height);

		for (cv::Point& p : m_maxContour)
			p += roi.tl();
		dst(roi).copyTo(pDib(rect));
	}
#ifdef LOG
	FileTools::write_log("imgprc.txt", "exit CImageApplyAutoCrop apply8");
#endif // LOG
}

void CImageApplyAutoCrop::apply(std::vector<cv::Mat>& mats, bool isTwoSide)
{
	if (mats.empty()) return;
	if (!mats[0].empty()) {
		apply(mats[0], 0);
		m_rects.push_back(m_rect);
		brightSharp(mats[0]);
	}

	if (isTwoSide && mats.size() > 1)
	{
		cv::Size dSize = m_fixedSize;
		if (!mats[0].empty())
			m_fixedSize = mats[0].size();
		if (!mats[1].empty()) {
			apply(mats[1], 1);
			m_rects.push_back(m_rect);
			brightSharp(mats[1]);
		}

		if (!mats[0].empty())
			m_fixedSize = dSize;
	}
}

cv::Scalar CImageApplyAutoCrop::getBackGroudColor(const cv::Mat& image, int total)
{
	if (image.channels() == 3)
	{
		cv::Mat image_bgr[3];
		cv::split(image, image_bgr);

		uchar bgr[3];
		for (size_t i = 0; i < 3; i++)
			bgr[i] = getBackGroudChannelMean(image_bgr[i], total);
		return cv::Scalar(bgr[0], bgr[1], bgr[2]);
	}
	else
		return cv::Scalar::all(getBackGroudChannelMean(image, total));
}

uchar CImageApplyAutoCrop::getBackGroudChannelMean(const cv::Mat& gray, int total)
{
	cv::Mat image_clone;
	cv::resize(gray, image_clone, cv::Size(), 0.25, 0.25);

	int threnshold = total / 32;
	int channels[] = { 0 };
	int nHistSize[] = { 256 };
	float range[] = { 0, 256 };
	const float* fHistRanges[] = { range };
	cv::Mat hist;
	cv::calcHist(&image_clone, 1, channels, cv::Mat(), hist, 1, nHistSize, fHistRanges, true, false);

	int hist_array[256];
	for (int i = 0; i < 256; i++)
		hist_array[i] = hist.at<float>(i, 0);

	int length = 1;
	const int length_max = 255 - m_threshold;
	while (length < length_max)
	{
		for (size_t i = m_threshold + 1; i < 256 - length; i++)
		{
			int count = 0;
			uint pixSum = 0;
			for (size_t j = 0; j < length; j++)
			{
				count += hist_array[j + i];
				pixSum += hist_array[j + i] * (i + j);
			}

			if (count >= threnshold)
				return pixSum / count;
		}
		length++;
	}
	return 255;
}