高效的图像遍历循环

在对图像进行操作时，免不了要对整个图像进行循环，下面对复制图像的三种不同的遍历方式进行比较

<pre name="code" class="cpp">int _tmain(int argc, _TCHAR* argv[])
{
	Mat src = imread("Lena.jpg", 0);
	if (src.empty())
	{
		cout << "failed" << endl;
		return -1;
	}
	imshow("src", src);

	// ......
	Mat dst(src.size(), src.type());
	double start = getTickCount();
	for (int r = 0; r < src.rows; r++)
	{
		const uchar* ps = src.ptr<uchar>(r);
		uchar* pd = dst.ptr<uchar>(r);
		for (int c = 0; c < src.cols; c++)
		{
			pd[c] = ps[c];
		}
	}
	double end = getTickCount();
	double time = (end - start) / getTickFrequency();
	cout << "dst1\t" << time << endl;
	imshow("dst1", dst);

	int rows = src.rows;
	int cols = src.cols;

	// ......
	dst.setTo(Scalar(0));
	start = getTickCount();
	if (src.isContinuous() && dst.isContinuous())
	{
		cols *= rows;
		rows = 1;
	}
	for (int r = 0; r < rows; r++)
	{
		const uchar* ps = src.ptr<uchar>(r);
		uchar* pd = dst.ptr<uchar>(r);
		for (int c = 0; c < cols; c++)
		{
			pd[c] = ps[c];
		}
	}
	end = getTickCount();
	time = (end - start) / getTickFrequency();
	cout << "dst2\t" << time << endl;
	imshow("dst2", dst);

	rows = src.rows;
	cols = src.cols;

	dst.setTo(Scalar(0));
	start = getTickCount();

	// ......
	// 此种遍历方式，在openCV源码中会经常看到
	if (src.isContinuous() && dst.isContinuous())
	{
		cols *= rows;
		rows = 1;
	}
	for (int r = 0; r < rows; r++)
	{
		const uchar* ps = src.ptr<uchar>(r);
		uchar* pd = dst.ptr<uchar>(r);
		int c = 0;
		for ( ; c < cols - 4; c += 4)
		{
			pd[c] = ps[c];
			pd[c + 1] = ps[c + 1];
			pd[c + 2] = ps[c + 2];
			pd[c + 3] = ps[c + 3];
		}

		for ( ; c < cols; c++)
		{
			pd[c] = ps[c];
		}
	}
	end = getTickCount();
	time = (end - start) / getTickFrequency();
	cout << "dst3\t" << time << endl;
	imshow("dst3", dst);

	waitKey();
	return 0;
}

下面是相对应的消耗的时间：

可以看出最后一种遍历方式是最快的。

方式2与方式1比较、方式3与方式2比较，存在一个for循环的优化上面，可以参考嵌套For循环性能优化案例。