yuv数据(nv12和nv21)和RGB数据之间转换的c++代码
yuv数据(以nv21和nv12为例)和RGB数据之间转换代码(c++)
一、首先了解下nv12和nv21的数据排布
nv21
Y Y Y Y
Y Y Y Y
Y Y Y Y
Y Y Y Y
V U V U
V U V U
nv21
Y Y Y Y
Y Y Y Y
Y Y Y Y
Y Y Y Y
U V U V
U V U V
主要就是UV的顺序不同,交互一下UV的位置就可以互换NV12和NV21.
二、bgr(rgb)转nv21(nv12)
一般手机等移动端的数据流格式都是yuv格式,而神经网络的输入一般都是rgb格式,所以需要进行转换,这里给出c++的代码示例。
cv::Mat bgr2yuv(cv::Mat &bgr)
{
cv::Mat img_yuv_yv12;
int height = bgr.rows;
int width = bgr.cols;
cv::Mat img_yuv(height * 3 / 2, width, CV_8UC1);
cv::cvtColor(bgr, img_yuv_yv12, CV_BGR2YUV_YV12);
memcpy(img_yuv.data, img_yuv_yv12.data, height * width);
char *v = (char*)img_yuv_yv12.data + height * width;
char *u = v + height * width / 4;
char *dst = (char*)img_yuv.data + height * width;
for (int i = 0;i < height * width / 4; ++i)
{
//nv21
dst[2 * i] = v[i];
dst[2 * i + 1] = u[i];
//nv12
//dst[2 * i] = u[i];
//dst[2 * i + 1] = v[i];
}
return img_yuv;
}
前面提到,交换UV的位置就可以得到NV21和NV12。
三、nv21(nv12)转BGR
int yuv2bgr(unsigned char * yuv_img, unsigned char *rgb_img,int width, int height)
{
unsigned char * ydata = yuv_img;
unsigned char *uvdata = ydata + width * height;
int indexY, indexU, indexV;
unsigned char Y, U, V;
int B, G, R;
for (int i = 0; i < height; i++)
{
for (int j = 0; j < width; j++)
{
indexY = i * width + j;
Y = ydata[indexY];
if (j % 2 == 0)
{
indexU = i / 2 * width + j;
indexV = indexU + 1;
U = uvdata[indexU];
V = uvdata[indexV];
}
else
{
indexV = i / 2 * width + j;
indexU = indexV - 1;
U = uvdata[indexU];
V = uvdata[indexV];
}
//nv21
R = (unsigned char)(Y + 1.4075 * (U - 128));
G = (unsigned char)(Y - 0.3455 * (V - 128) - 0.7169 * (U - 128));
B = (unsigned char)(Y + 1.779 * (V - 128));
//nv12
//R = (unsigned char)(Y + 1.4075 * (V - 128));
//G = (unsigned char)(Y - 0.3455 * (U - 128) - 0.7169 * (V - 128));
//B = (unsigned char)(Y + 1.779 * (U - 128));
rgb_img[indexY * 3 + 0] = clamp_g(B, 0, 255);
rgb_img[indexY * 3 + 1] = clamp_g(G, 0, 255);
rgb_img[indexY * 3 + 2] = clamp_g(R, 0, 255);
}
}
return 0;
}
四、效果

OK!
五、速度优化
nv21转bgr的平均时间(输入大小720p):
AveTime:13.7835ms
改进代码:
改进后平均时间:AveTime:5.52088ms
速度提升60%,结果完全一致。
更多推荐
所有评论(0)