实现彩色空间转换 即RGB到YUV240 和 YUV240到RGB的转换
一. 彩色空间转换的基本公式
1. RGB转YUV
亮度和色差信号:
- Y=0.2990R+0.5870G+0.1140B
- R-Y=0.7010R-0.5870G-0.1140B
- B-Y=-0.2990R-0.5870G+0.8860B
为了使色差信号的动态范围控制在0.5之间,需要进行归一化,对色差信号引入压缩系数。归一化后的色差信号为:
- U=-0.1684R-0.3316G+0.5B
- V=0.5R-0.4187G-0.0813B
2. YUV转RGB
同理,我们可以逆向得出YUV转换到RGB的公式:
- R = Y + 1.402 (V - 128)
- G = Y - 0.34413 (U - 128) - 0.71414(V - 128)
- B=Y + 1.772*(U - 128)
二. 使用C++ 实现RGB到YUV240的转换
1. 打开RGB文件与创建待存储的YUV文件
我们使用fopen函数打开和创建文件,在c++中如果遇到fopen报错的情况,我们可以在 调试->调试属性->C/C++预处理器->预处理器定义 中添加一句_CRT_SECURE_NO_WARNINGS
即可正常使用fopen了。
/* open the RGB file */
rgbFile = fopen(rgbFileName, "rb");
if (rgbFile == NULL)
{
printf("cannot find rgb file\n");
exit(1);
}
else
{
printf("The input rgb file is %s\n", rgbFileName);
}
/* open the RAW file */
yuvFile = fopen(yuvFileName, "wb");
if (yuvFile == NULL)
{
printf("cannot find yuv file\n");
exit(1);
}
else
{
printf("The output yuv file is %s\n", yuvFileName);
}
2. 创建存储区Buffer
/* get an input buffer for a frame */
rgbBuf = (u_int8_t*)malloc(frameWidth * frameHeight * 3);
/* get the output buffers for a frame */
yBuf = (u_int8_t*)malloc(frameWidth * frameHeight);
uBuf = (u_int8_t*)malloc((frameWidth * frameHeight) / 4);
vBuf = (u_int8_t*)malloc((frameWidth * frameHeight) / 4);
if (rgbBuf == NULL || yBuf == NULL || uBuf == NULL || vBuf == NULL)
{
printf("no enought memory\n");
exit(1);
}
3. 主函数以及码电平分配及数字表达式
- 亮电平信号量化后码电平分配:在对分量信号进行8比特均匀量化时,共分为256个等间隔的量化级。为了防止信号变动造成过载,在256级上端留20级,下端留16级作为信号超越动态范围的保护带。
- 色差信号量化后码电平分配:色差信号经过归一化处理后,动态范围为-0.5-0.5,让色差零电平对应码电平128,色差信号总共占225个量化级。在256级上端留15级,下端留16级作为信号超越动态范围的保护带。
while (fread(rgbBuf, 1, frameWidth * frameHeight * 3, rgbFile))
{
if(RGB2YUV(frameWidth, frameHeight, rgbBuf, yBuf, uBuf, vBuf, flip))
{
printf("error");
return 0;
}
for (i = 0; i < frameWidth*frameHeight; i++)
{
if (yBuf[i] < 16) yBuf[i] = 16;
if (yBuf[i] > 235) yBuf[i] = 235;
}
for (i = 0; i < frameWidth*frameHeight/4; i++)
{
if (uBuf[i] < 16) uBuf[i] = 16;
if (uBuf[i] > 240) uBuf[i] = 240;
if (vBuf[i] < 16) vBuf[i] = 16;
if (vBuf[i] > 240) vBuf[i] = 240;
}
fwrite(yBuf, 1, frameWidth * frameHeight, yuvFile);
fwrite(uBuf, 1, (frameWidth * frameHeight) / 4, yuvFile);
fwrite(vBuf, 1, (frameWidth * frameHeight) / 4, yuvFile);
printf("\r...%d", ++videoFramesWritten);
}
4. rgb2yuv函数的编写
- 内存分配
y_buffer = (unsigned char *)y_out;
sub_u_buf = (unsigned char *)u_out;
sub_v_buf = (unsigned char *)v_out;
u_buffer = (unsigned char *)malloc(size * sizeof(unsigned char));
v_buffer = (unsigned char *)malloc(size * sizeof(unsigned char));
if (!(u_buffer && v_buffer))
{
if (u_buffer) free(u_buffer);
if (v_buffer) free(v_buffer);
return 2;
}
b = (unsigned char *)bmp;
y = y_buffer;
u = u_buffer;
v = v_buffer;
- 将RGB转换为YUV
if (!flip) {
for (j = 0; j < y_dim; j ++)
{
y = y_buffer + (y_dim - j - 1) * x_dim;
u = u_buffer + (y_dim - j - 1) * x_dim;
v = v_buffer + (y_dim - j - 1) * x_dim;
for (i = 0; i < x_dim; i ++) {
g = b + 1;
r = b + 2;
*y = (unsigned char)( RGBYUV02990[*r] + RGBYUV05870[*g] + RGBYUV01140[*b]);
*u = (unsigned char)(- RGBYUV01684[*r] - RGBYUV03316[*g] + (*b)/2 + 128);
*v = (unsigned char)( (*r)/2 - RGBYUV04187[*g] - RGBYUV00813[*b] + 128);
b += 3;
y ++;
u ++;
v ++;
}
}
} else {
for (i = 0; i < size; i++)
{
g = b + 1;
r = b + 2;
*y = (unsigned char)( RGBYUV02990[*r] + RGBYUV05870[*g] + RGBYUV01140[*b]);
*u = (unsigned char)(- RGBYUV01684[*r] - RGBYUV03316[*g] + (*b)/2 + 128);
*v = (unsigned char)( (*r)/2 - RGBYUV04187[*g] - RGBYUV00813[*b] + 128);
b += 3;
y ++;
u ++;
v ++;
}
}
- 对UV子样本的调整
for (j = 0; j < y_dim/2; j ++)
{
psu = sub_u_buf + j * x_dim / 2;
psv = sub_v_buf + j * x_dim / 2;
pu1 = u_buffer + 2 * j * x_dim;
pu2 = u_buffer + (2 * j + 1) * x_dim;
pv1 = v_buffer + 2 * j * x_dim;
pv2 = v_buffer + (2 * j + 1) * x_dim;
for (i = 0; i < x_dim/2; i ++)
{
*psu = (*pu1 + *(pu1+1) + *pu2 + *(pu2+1)) / 4;
*psv = (*pv1 + *(pv1+1) + *pv2 + *(pv2+1)) / 4;
psu ++;
psv ++;
pu1 += 2;
pu2 += 2;
pv1 += 2;
pv2 += 2;
}
}
5. YUV文件的输出对比
右侧的dowm+new是输出的yuv图片
三. 使用Python 实现YUV240到RGB的转换
下面展示该程序的主要功能代码部分:
1. YUV通道的分离与扩展
在这里首先是YUV三通道的分离,因为本次实验采用的文件分辨率为256256,且该二进制图像文件的存储格式为:yuv格式按照全部像素的y数据块、u数据块、v数据块以此存放,即“YYYY… UUUU… VVVV…”
然后是UV二通道的扩展,因为YUV是4:2:0格式的,所以UV的数据均变成了原来的四分之一,所以我通过两层循环将UV也扩展为256256的大小。
#YUV通道的分离与扩展
data_y = []
uGroup = []
vGroup = []
for i in range(256*256):
data_y.append(data[i])
for i in range(256*256, int(256*256*1.25)):
uGroup.append(data[i])
for i in range(int(256*256*1.25), int(256*256*1.5)):
vGroup.append(data[i])
data_u = []
for i in range(128):
for j in range(128):
data_u.append(uGroup[j*i])
data_u.append(uGroup[j*i])
for m in range(128):
data_u.append(uGroup[m*i])
data_u.append(uGroup[m*i])
data_v = []
for i in range(128):
for j in range(128):
data_v.append(vGroup[j*i])
data_v.append(vGroup[j*i])
for m in range(128):
data_v.append(vGroup[m*i])
data_v.append(vGroup[m*i])
2. YUV向RGB的转换
在这一步中我们就使用了,一开始得到的转换公式,同时因为RGB三通道的取值范围是[0,255],所以我们要对超出这一范围的非法值进行限制。最后在按照rgb文件按照每个像素BGR分量一次储存,即“BGR BGR BGR …”的格式进行三通道的合并。
rGroup = []
gGroup = []
bGroup = []
data_rgb = []
for i in range(256*256):
rGroup.append(data_y[i] + 1.402 * (data_v[i] - 128))
gGroup.append(data_y[i] - 0.34413 * (data_u[i] - 128) - 0.71414*(data_v[i] - 128))
bGroup.append(data_y[i] + 1.772*(data_u[i] - 128))
for i in range(256*256):
rGroup[i] = 0 if rGroup[i] < 0 else rGroup[i]
gGroup[i] = 0 if gGroup[i] < 0 else gGroup[i]
bGroup[i] = 0 if bGroup[i] < 0 else bGroup[i]
rGroup[i] = 255 if rGroup[i] > 255 else rGroup[i]
gGroup[i] = 255 if gGroup[i] > 255 else gGroup[i]
bGroup[i] = 255 if bGroup[i] > 255 else bGroup[i]
for i in range(256*256):
data_rgb.append(bGroup[i])
data_rgb.append(gGroup[i])
data_rgb.append(rGroup[i])
3. YUV向RGB的转换后的图像与原来RGB图像的对比
左边是我们原来的RGB图像,右面是我们通过YUV向RGB的转换后的图像。可以观察到,经过转换后的图像的色彩深度和亮度层次都不如原来的图像。其中的原因也不难得出。因为原图在转换到YUV是4:2:0的采样,本身就丢失了很多色度信息,在相互转换的过程中,公式中的系数也是有限小数位的,也达不到完全转换的精度
4. 两张图像RGB三通道的概率分布的对比
左边是我们原来的RGB图像RGB三通道的概率分布,右面是我们通过YUV向RGB的转换后的图像RGB三通道的概率分布。很明显就可以看出,后来的图像的色彩要弱于原图像
5. 完整代码
#导入程序所需要的包
import cv2 as cv
import numpy as np
import math as ma
import matplotlib.pyplot as plt
#以2进制文件读取"down.yuv"文件
f = open("down.yuv", "rb")
f2 = open("down.rgb", "rb")
data = f.read()
data2 = f2.read()
#关闭打开的文件
f.close()
f2.close()
#将数据转换成int
data = [int(x) for x in data]
data2 = [int(x) for x in data2]
#YUV通道的分离与扩展
data_y = []
uGroup = []
vGroup = []
for i in range(256*256):
data_y.append(data[i])
for i in range(256*256, int(256*256*1.25)):
uGroup.append(data[i])
for i in range(int(256*256*1.25), int(256*256*1.5)):
vGroup.append(data[i])
data_u = []
for i in range(128):
for j in range(128):
data_u.append(uGroup[j*i])
data_u.append(uGroup[j*i])
for m in range(128):
data_u.append(uGroup[m*i])
data_u.append(uGroup[m*i])
data_v = []
for i in range(128):
for j in range(128):
data_v.append(vGroup[j*i])
data_v.append(vGroup[j*i])
for m in range(128):
data_v.append(vGroup[m*i])
data_v.append(vGroup[m*i])
#YUV向RGB的转换
rGroup = []
gGroup = []
bGroup = []
data_rgb = []
for i in range(256*256):
rGroup.append(data_y[i] + 1.402 * (data_v[i] - 128))
gGroup.append(data_y[i] - 0.34413 * (data_u[i] - 128) - 0.71414*(data_v[i] - 128))
bGroup.append(data_y[i] + 1.772*(data_u[i] - 128))
for i in range(256*256):
rGroup[i] = 0 if rGroup[i] < 0 else rGroup[i]
gGroup[i] = 0 if gGroup[i] < 0 else gGroup[i]
bGroup[i] = 0 if bGroup[i] < 0 else bGroup[i]
rGroup[i] = 255 if rGroup[i] > 255 else rGroup[i]
gGroup[i] = 255 if gGroup[i] > 255 else gGroup[i]
bGroup[i] = 255 if bGroup[i] > 255 else bGroup[i]
for i in range(256*256):
data_rgb.append(bGroup[i])
data_rgb.append(gGroup[i])
data_rgb.append(rGroup[i])
#转换格式使用OpenCV显示
#原始的rgb图像
data2 = np.array(data2).reshape((256, 256, 3)).astype(np.uint8)
cv.imshow("down.rgb", data2)
#通过yuv转换成rgb的图像
data_rgb = np.array(data_rgb).reshape((256, 256, 3)).astype(np.uint8)
cv.imshow("down_yuv2rgb.rgb", data_rgb)
cv.waitKey()
#计算三通道的概率分布
#b通道
b_gailv = [0 for x in range(0, 256)]#创建一个长度为256的默认值为0的列表
b_gailv = [int(x) for x in b_gailv]
for i in range(256*256):
for j in range(256):
if bGroup[i]==j:
b_gailv[j] = b_gailv[j] + 1
for j in range(256):
b_gailv[j] = b_gailv[j]/(256*256)
#g通道
g_gailv = [0 for x in range(0, 256)]#创建一个长度为256的默认值为0的列表
g_gailv = [int(x) for x in g_gailv]
for i in range(256*256):
for j in range(256):
if gGroup[i]==j:
g_gailv[j] = g_gailv[j] + 1
for j in range(256):
g_gailv[j] = g_gailv[j]/(256*256)
#r通道
r_gailv = [0 for x in range(0, 256)]#创建一个长度为256的默认值为0的列表
r_gailv = [int(x) for x in r_gailv]
for i in range(256*256):
for j in range(256):
if rGroup[i]==j:
r_gailv[j] = r_gailv[j] + 1
for j in range(256):
r_gailv[j] = r_gailv[j]/(256*256)
#作图
plt.plot(b_gailv, 'b')
plt.plot(g_gailv, 'g')
plt.plot(r_gailv, 'r')
plt.legend(["b aisle", "g aisle", "r aisle"])
plt.show()
四. 归纳总结
为了实现彩色空间的转换,我们要熟练掌握RGB与YUV之间的转换公式,并且熟悉YUV的采样格式:由左到右分别是4:4:4,4:2:2,4:2:0(X代表亮度采样,O代表色度取样)
在编写C++程序中要尤其注意内存的分配和使用,在两种语言的编写中都要注意RGB和YUV的取值范围,以免生成不合法的数据造成报错。