Question

我正在尝试在一个YUV格式的OpenCV图像上进行图像稳定，其中有2个通道：一个包含Y值，另一个包含交替的U / V值（OpenCV中的YUV_YUY2）。

我遇到的问题是某些矩阵运算，特别是warpAffine和resize，可以移动数组，使得U值最终在V位置，反之亦然，反转颜色或在图像中引起条纹。 / p>

我的第一个想法是我可以将图像转换为不同的（非交错的）图像类型，应用转换，然后转换回来，但是使用我发现的cvtColor（）的可用转换：http://docs.opencv.org/3.1.0/d7/d1b/group__imgproc__misc.html

我可以从YUV_YUY2转换为其他格式，但没有可用的转换回YUV_YUY2格式。

我或许可以手动重建YUY2矩阵，但我怀疑这可能是一种我更容易丢失的方法。任何OpenCV大师有什么建议吗？感谢。

代码段：

cv::Mat stabilize (VideoFrame* curFrame, VideoFrame* prevFrame)
{
   Mat locCurMat, locPrevMat;
   locCurMat = Mat(1080, 1920, CV_8UC2, curFrame->video);
   locPrevMat = Mat(1080, 1920, CV_8UC2, prevFrame->video);

   // Calculate the transformation matrix
   Mat locTransform;
   // I’ll spare you this part
   // …

   // Apply the transformation
   Mat locStabFrame;
   // Make the border value (0, 128) for black: 
   // If you leave it at 0,0 it's bright green
   Scalar locBorderVal(0, 128);
   warpAffine(locCurMat, locStabFrame, locTransform, locCurMat.size(),
              INTER_LINEAR, BORDER_CONSTANT, locBorderVal);

   // Crop the borders to make the video look better
   // Get the aspect ratio correct
   int locVertBorder = HORIZONTAL_BORDER_CROP * 
                       locCurMat.rows / locCurMat.cols;
   locStabFrame = locStabFrame(
       Range(locVertBorder, locStabFrame.rows - locVertBorder),
       Range(HORIZONTAL_BORDER_CROP,
             locStabFrame.cols - HORIZONTAL_BORDER_CROP));

   // Resize locStabFrame back to the proper frame size
   resize(locStabFrame, locStabFrame, locCurMat.size());

   // Return the stabilized result
   return locStabFrame;
}

Answer 1

如果有人有兴趣，这是我为避免这个问题而采取的流程。但是它很流氓，如果有人有更高效的方式，我会很高兴听到它。

步骤基本上是：

在2通道YUV
从2通道YUV转换为RGB
运行warpAffine＆amp;调整RGB图像大小
将RGB转换为3通道YUV
使用u / v掩码从3通道YUV重建2通道YUV
返回2频道YUV图像

代码：

// Apply the transformation
Mat rgbFrame, stabFrame, outputFrame;
// Convert the frame to RGB
cvtColor(locCurMat, rgbFrame, COLOR_YUV2RGB_YUY2);

// Apply the image translation
warpAffine(rgbFrame, stabFrame, T, rgbFrame.size());

// Crop the borders to make the video look better
// Get the aspect ratio correct
int vert_border = HORIZONTAL_BORDER_CROP * locCurMat.rows / locCurMat.cols;
stabFrame = stabFrame(Range(vert_border, stabFrame.rows - vert_border),
                      Range(HORIZONTAL_BORDER_CROP,
                            stabFrame.cols - HORIZONTAL_BORDER_CROP));


// Resize regFrame back to the correct frame size
resize(stabFrame, stabFrame, rgbFrame.size());

// Convert back from RGB to YUV
// First, see if the u/v masks are the correct size, and populate them
// if not. Hopefully this only happens once
if (locCurMat.rows != m_uMask.rows || locCurMat.cols != m_uMask.cols)
{
     printf("Creating U/V image masks\n");
     m_uMask = cv::Mat::zeros(locCurMat.size(), CV_8UC1);
     m_vMask = cv::Mat::zeros(locCurMat.size(), CV_8UC1);
     cv::Mat onesCol = cv::Mat::ones(locCurMat.rows, 1, CV_8UC1);
     // Populate the masks with alternating columns of 1s
     for (int i = 0; i < locCurMat.cols; i++)
     {
         // Copy even columns to the u mask
         if (i % 2 == 0)
         {
             onesCol.copyTo(m_uMask.col(i));
         }
         // Copy odd columns to the v mask
         else
         {
             onesCol.copyTo(m_vMask.col(i));
         }
      }
  }

// Convert the stabilized rgb frame to YUV
Mat yuvFrame;
cvtColor(stabFrame, yuvFrame, COLOR_RGB2YUV);

// Split the YUV frame into individual channels
vector<Mat> yuv3Channels;
split(yuvFrame, yuv3Channels);

// Copy the 3 channel YUV values into 2 channel YUV
vector<Mat> yuv2Channels;
Mat uvChannel;
yuv3Channels[1].copyTo(uvChannel, m_uMask);
yuv3Channels[2].copyTo(uvChannel, m_vMask);
yuv2Channels.push_back(yuv3Channels[0]);
yuv2Channels.push_back(uvChannel);

// Merge the 2 channel yuv back into the output frame
cv::merge(yuv2Channels, outputFrame);

OpenCV矩阵操作导致YUV图像的颜色损坏

1 个答案: