Revision history [back]

potential loss of information in processing 16bit YUV greyscale

Ive been looking at cap_v4l.cpp

specifically

case V4L2_PIX_FMT_Y16:
{
    cv::Mat temp(imageSize, CV_8UC1, buffers[MAX_V4L_BUFFERS].start);
    cv::Mat(imageSize, CV_16UC1, currentBuffer.start).convertTo(temp, CV_8U, 1.0 / 256);
    cv::cvtColor(temp, destination, COLOR_GRAY2BGR);
    return;
}
case V4L2_PIX_FMT_Y12:
{
    cv::Mat temp(imageSize, CV_8UC1, buffers[MAX_V4L_BUFFERS].start);
    cv::Mat(imageSize, CV_16UC1, currentBuffer.start).convertTo(temp, CV_8U, 1.0 / 16);
    cv::cvtColor(temp, destination, COLOR_GRAY2BGR);
    return;
}
case V4L2_PIX_FMT_Y10:
{
    cv::Mat temp(imageSize, CV_8UC1, buffers[MAX_V4L_BUFFERS].start);
    cv::Mat(imageSize, CV_16UC1, currentBuffer.start).convertTo(temp, CV_8U, 1.0 / 4);
    cv::cvtColor(temp, destination, COLOR_GRAY2BGR);
    return;
}

to me this looks like it converts 16bit (12bit or 10bit) to 8bit before converting to BGR an would therefore lose accuracy. am i correct in thinking this? should this code be using CV_16U instead?

potential loss of information in processing 16bit YUV greyscale

Ive been looking at cap_v4l.cpp

specifically

case V4L2_PIX_FMT_Y16:
{
    cv::Mat temp(imageSize, CV_8UC1, buffers[MAX_V4L_BUFFERS].start);
    cv::Mat(imageSize, CV_16UC1, currentBuffer.start).convertTo(temp, CV_8U, 1.0 / 256);
    cv::cvtColor(temp, destination, COLOR_GRAY2BGR);
    return;
}
case V4L2_PIX_FMT_Y12:
{
    cv::Mat temp(imageSize, CV_8UC1, buffers[MAX_V4L_BUFFERS].start);
    cv::Mat(imageSize, CV_16UC1, currentBuffer.start).convertTo(temp, CV_8U, 1.0 / 16);
    cv::cvtColor(temp, destination, COLOR_GRAY2BGR);
    return;
}
case V4L2_PIX_FMT_Y10:
{
    cv::Mat temp(imageSize, CV_8UC1, buffers[MAX_V4L_BUFFERS].start);
    cv::Mat(imageSize, CV_16UC1, currentBuffer.start).convertTo(temp, CV_8U, 1.0 / 4);
    cv::cvtColor(temp, destination, COLOR_GRAY2BGR);
    return;
}

to me this looks like it converts 16bit (12bit or 10bit) to 8bit before converting to BGR an would therefore lose accuracy. am i correct in thinking this? should this code be using CV_16U instead?