YUV420 to BGR image from pixel pointers

Question

I am capturing raw output from a decoder which is YUV420. I have got three pointers: Y(1920*1080), U(960*540) and V(960*540) separately.

I want to save the image as JPEG using OpenCV. I tried using cvtcolor of opencv

cv::Mat i_image(cv::Size(columns, rows), CV_8UC3, dataBuffer);
cv::Mat i_image_BGR(cv::Size(columns, rows), CV_8UC3);
cvtColor(i_image, i_image_BGR, cv::COLOR_YCrCb2BGR);
cv::imwrite("/data/data/org.myproject.debug/files/pic1.jpg", i_image_BGR);

But, here is the output image which is saved:

Can someone please suggest what is the proper way of saving the image?

YUV Binary files for reference

Can you share the complete code? Seems like you are not converting YUV420 to YUV444 properly. — sgarizvi, May 08 '18 at 09:34
we directly tried to convert from YUV420 to BGR. the above snippet is the whole thing used inside a function. Can you please share more information on how to convert YUV420 to YUV444 ? — kakumanu-sudhir, May 08 '18 at 09:40
You said you've 3 pointers but you're only providing one pointer to `i_image`. — zindarod, May 08 '18 at 09:52
It looks like the U and V components have to be replicated for 4 Y components. If you can share the individual input images (or the values of `dataBuffer`), I will be able to test a sample code. — sgarizvi, May 08 '18 at 09:52
@zindarod... Yeah, I think OP is mis-interpreting the pixel values in the `dataBuffer`. — sgarizvi, May 08 '18 at 09:53
@zindarod... dataBuffer is the Y(1920*1080) plane[0] output of decoder. I am not sure how to utilize the U(960*540) plane[1] and V(960*540) plane[2] in OpenCV. — kakumanu-sudhir, May 08 '18 at 10:01
@sgarizvi since i have three data buffers, is it okay if i can share the data in binary format files and share it to you? — kakumanu-sudhir, May 08 '18 at 10:03
Read this [page](https://en.wikipedia.org/wiki/YUV#Y%E2%80%B2UV420p_(and_Y%E2%80%B2V12_or_YV12)_to_RGB888_conversion), notice the total size of buffer, the position of Y, U and V components. Create a new buffer and copy the data from the 3 pointers to the correct positions in the new pointer. — zindarod, May 08 '18 at 10:09
@sgarizvi Yeah, he just needs to understand the YUV data layout. — zindarod, May 08 '18 at 10:09
@sgarizvi, i have edited the post with link for uploaded YUV binary files. thanks. — kakumanu-sudhir, May 08 '18 at 12:13

sgarizvi · Accepted Answer · 2018-05-28T09:14:01.733

Following is the process to convert the provided YUV files into RGB image.

Read Y, U and V binary files into byte buffers
Create OpenCV Mat object from the created buffers.
Resize U and V Mats to the size of Y.
Merge Y and resized U and V.
Convert from YUV to BGR

Be advised that the resizing step is just an optimized way of repeating the values of U and V. This is only valid in the case where Y has twice the resolution of U and V in both dimensions. This approach should be invalid for arbitrary size images (not tested).

Here is the code for the above-mentioned process.

#include <iostream>
#include <vector>
#include <opencv2/opencv.hpp>

std::vector<unsigned char> readBytesFromFile(const char* filename)
{
    std::vector<unsigned char> result;

    FILE* f = fopen(filename, "rb");

    fseek(f, 0, SEEK_END);  // Jump to the end of the file
    long length = ftell(f); // Get the current byte offset in the file
    rewind(f);              // Jump back to the beginning of the file

    result.resize(length);

    char* ptr = reinterpret_cast<char*>(&(result[0]));
    fread(ptr, length, 1, f); // Read in the entire file
    fclose(f); // Close the file

    return result;
}

int main(int argc, char** argv)
{
    cv::Size actual_size(1920, 1080);
    cv::Size half_size(960, 540);

    //Read y, u and v in bytes arrays
    auto y_buffer = readBytesFromFile("ypixel.bin");
    auto u_buffer = readBytesFromFile("upixel.bin");
    auto v_buffer = readBytesFromFile("vpixel.bin");


    cv::Mat y(actual_size, CV_8UC1, y_buffer.data());
    cv::Mat u(half_size, CV_8UC1, u_buffer.data());
    cv::Mat v(half_size, CV_8UC1, v_buffer.data());

    cv::Mat u_resized, v_resized;
    cv::resize(u, u_resized, actual_size, 0, 0, cv::INTER_NEAREST); //repeat u values 4 times
    cv::resize(v, v_resized, actual_size, 0, 0, cv::INTER_NEAREST); //repeat v values 4 times

    cv::Mat yuv;

    std::vector<cv::Mat> yuv_channels = { y, u_resized, v_resized };
    cv::merge(yuv_channels, yuv);

    cv::Mat bgr;
    cv::cvtColor(yuv, bgr, cv::COLOR_YUV2BGR);
    cv::imwrite("bgr.jpg", bgr);

    return 0;
}

Compiled and tested with following command:

g++ -o yuv2rgb -std=c++11 yuv2rgb.cpp -L/usr/local/lib -lopencv_core -lopencv_imgcodecs -lopencv_highgui -lopencv_imgproc

Following output image is generated by executing the above code:

Works like a charm. Thanks a lot sgarizvi. – kakumanu-sudhir May 08 '18 at 13:18 — kakumanu-sudhir, May 08 '18 at 13:18

score 2 · Answer 2 · answered May 08 '18 at 10:30

2

I think OpenCV matrix for your input yuv420 planar image should have 1 channel format instead of 3 channel. Place there Y channel, then U, then V. I found a very similar question HERE Planar YUV420 and NV12 is the same

answered May 08 '18 at 10:30

VideoProcessingResearcher

498
3
16

YUV420 to BGR image from pixel pointers

2 Answers2

Linked