我想要一对转换算法，一个从 RGB 到 YUV，另一个从 YUV 到 RGB，它们是互逆的。也就是说，往返转换应该保持该值不变。（如果愿意，可以将 YUV 替换为 Y'UV、YUV、YCbCr、YPbPr。）
这样的事情存在吗？如果是这样，它是什么？
发布的解决方案（如何在 C/C++ 中执行 RGB->YUV 转换？，http://www.fourcc.org/fccyvrgb.php，http : //en.wikipedia.org/wiki/YUV）只是逆向的（当省略对 [0,255] 的钳位时，两个 3x3 矩阵是逆矩阵。但是省略该钳位会允许出现负亮度之类的事情，这会对 YUV 空间中的图像处理造成严重破坏。保留钳位会使转换非线性，这使得定义逆变得棘手。
是的，存在可逆变换。
equasys GmbH发布了从 RGB 到 YUV、YCbCr 和 YPbPr 的可逆转换，并解释了每种情况适用于哪种情况、这种钳制的真正含义以及参考链接。（就像一个很好的答案。）
对于我自己的应用程序（jpg 图像，而不是模拟电压）YCbCr 是合适的，所以我为这两个转换编写了代码。事实上，对于许多图像，在 256 中反复出现的值相差不到 1 分；前后图像在视觉上无法区分。
PIL 的色彩空间转换 YCbCr -> RGB因提及 equasys 的网页而受到赞誉。
其他答案，可能会提高 equasys 的精度和简洁性：
https://code.google.com/p/imagestack/包括 rgb_to_x 和 x_to_rgb 函数，但我没有尝试编译和测试它们。
Cory Nelson 的答案链接到具有类似功能的代码，但它说一般情况下不可能进行反转，这与 equasys 相矛盾。
FFmpeg、OpenCV、VLFeat 或 ImageMagick 的源代码。
2019 年编辑： 这是来自 github 的 C 代码，在我的评论中提到。
void YUVfromRGB(double& Y, double& U, double& V, const double R, const double G, const double B)
{
Y = 0.257 * R + 0.504 * G + 0.098 * B + 16;
U = -0.148 * R - 0.291 * G + 0.439 * B + 128;
V = 0.439 * R - 0.368 * G - 0.071 * B + 128;
}
void RGBfromYUV(double& R, double& G, double& B, double Y, double U, double V)
{
Y -= 16;
U -= 128;
V -= 128;
R = 1.164 * Y + 1.596 * V;
G = 1.164 * Y - 0.392 * U - 0.813 * V;
B = 1.164 * Y + 2.017 * U;
}
维基百科上有一个关于YUV主题的漂亮图表，它描述了 YUV420p 的布局。但是，如果您像我一样想要 NV21，有时称为 YUV420sp，它将 V 和 U 分量交错在一个平面中，因此在这种情况下，该图是错误的，但它可以让您直观地了解它是如何工作的。
此格式 (NV21) 是 Android 相机预览上的标准图片格式。YUV 4:2:0 平面图像，具有 8 位 Y 样本，然后是具有 8 位 2x2 子采样色度样本的交错 V/U 平面。
所以我看到的很多代码只是开始按照这个规范进行编码，而不考虑Endianess。此外，它们往往只支持 YUV 到 RGB 并且只支持一种或两种格式。然而，我想要一些更值得信赖的东西，结果证明从 Android 源代码存储库中获取的 C++ 代码可以解决问题。它几乎是直接的 C++，应该很容易在任何项目中使用。
在这种情况下，从 Java，但很容易使用 C 或 C++，您传入一个包含 RGB565 图像的字节数组并输出一个 NV21 字节数组。
#include <jni.h>
#include <cstring>
#include <cstdint>
#include "Converters.h"
#define JNI(X) JNIEXPORT Java_algorithm_ImageConverter_##X
#ifdef __cplusplus
extern "C" {
#endif
void JNI(RGB565ToNV21)(JNIEnv *env, jclass *, jbyteArray aRGB565in, jbyteArray aYUVout, jint width, jint height) {
//get jbyte array into C space from JVN
jbyte *rgb565Pixels = env->GetByteArrayElements(aRGB565in, NULL);
jbyte *yuv420sp = env->GetByteArrayElements(aYUVout, NULL);
size_t pixelCount = width * height;
uint16_t *rgb = (uint16_t *) rgb565Pixels;
// This format (NV21) is the standard picture format on Android camera preview. YUV 4:2:0 planar
// image, with 8 bit Y samples, followed by interleaved V/U plane with 8bit 2x2 subsampled
// chroma samples.
int uvIndex = pixelCount;
for (int row = 0; row < height; row++) {
for (int column = 0; column < width; column++) {
int pixelIndex = row * width + column;
uint8_t y = 0;
uint8_t u = 0;
uint8_t v = 0;
chroma::RGB565ToYUV(rgb[pixelIndex], &y, &u, &v);
yuv420sp[pixelIndex] = y;
if (row % 2 == 0 && pixelIndex % 2 == 0) {
#if __BYTE_ORDER == __LITTLE_ENDIAN
yuv420sp[uvIndex++] = u;
yuv420sp[uvIndex++] = v;
#else
yuv420sp[uvIndex++] = v;
yuv420sp[uvIndex++] = u;
#endif
}
}
}
//release temp reference of jbyte array
env->ReleaseByteArrayElements(aYUVout, yuv420sp, 0);
env->ReleaseByteArrayElements(aRGB565in, rgb565Pixels, 0);
}
#ifdef __cplusplus
}
#endif
正如您将在标题中看到的那样，有许多不同的转换选项可用于任意数量的格式。
/*
* Copyright (C) 2011 The Android Open Source Project
*
* Licensed under the Apache License, Version 2.0 (the "License");
* you may not use this file except in compliance with the License.
* You may obtain a copy of the License at
*
* http://www.apache.org/licenses/LICENSE-2.0
*
* Unless required by applicable law or agreed to in writing, software
* distributed under the License is distributed on an "AS IS" BASIS,
* WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
* See the License for the specific language governing permissions and
* limitations under the License.
*/
#ifndef HW_EMULATOR_CAMERA_CONVERTERS_H
#define HW_EMULATOR_CAMERA_CONVERTERS_H
#include <endian.h>
#ifndef __BYTE_ORDER
#error "could not determine byte order"
#endif
/*
* Contains declaration of framebuffer conversion routines.
*
* NOTE: RGB and big/little endian considerations. Wherever in this code RGB
* pixels are represented as WORD, or DWORD, the color order inside the
* WORD / DWORD matches the one that would occur if that WORD / DWORD would have
* been read from the typecasted framebuffer:
*
* const uint32_t rgb = *reinterpret_cast<const uint32_t*>(framebuffer);
*
* So, if this code runs on the little endian CPU, red color in 'rgb' would be
* masked as 0x000000ff, and blue color would be masked as 0x00ff0000, while if
* the code runs on a big endian CPU, the red color in 'rgb' would be masked as
* 0xff000000, and blue color would be masked as 0x0000ff00,
*/
namespace chroma {
/*
* RGB565 color masks
*/
#if __BYTE_ORDER == __LITTLE_ENDIAN
static const uint16_t kRed5 = 0x001f;
static const uint16_t kGreen6 = 0x07e0;
static const uint16_t kBlue5 = 0xf800;
#else // __BYTE_ORDER
static const uint16_t kRed5 = 0xf800;
static const uint16_t kGreen6 = 0x07e0;
static const uint16_t kBlue5 = 0x001f;
#endif // __BYTE_ORDER
static const uint32_t kBlack16 = 0x0000;
static const uint32_t kWhite16 = kRed5 | kGreen6 | kBlue5;
/*
* RGB32 color masks
*/
#if __BYTE_ORDER == __LITTLE_ENDIAN
static const uint32_t kRed8 = 0x000000ff;
static const uint32_t kGreen8 = 0x0000ff00;
static const uint32_t kBlue8 = 0x00ff0000;
#else // __BYTE_ORDER
static const uint32_t kRed8 = 0x00ff0000;
static const uint32_t kGreen8 = 0x0000ff00;
static const uint32_t kBlue8 = 0x000000ff;
#endif // __BYTE_ORDER
static const uint32_t kBlack32 = 0x00000000;
static const uint32_t kWhite32 = kRed8 | kGreen8 | kBlue8;
/*
* Extracting, and saving color bytes from / to WORD / DWORD RGB.
*/
#if __BYTE_ORDER == __LITTLE_ENDIAN
/* Extract red, green, and blue bytes from RGB565 word. */
#define R16(rgb) static_cast<uint8_t>((rgb) & kRed5)
#define G16(rgb) static_cast<uint8_t>(((rgb) & kGreen6) >> 5)
#define B16(rgb) static_cast<uint8_t>(((rgb) & kBlue5) >> 11)
/* Make 8 bits red, green, and blue, extracted from RGB565 word. */
#define R16_32(rgb) static_cast<uint8_t>((((rgb) & kRed5) << 3) | (((rgb) & kRed5) >> 2))
#define G16_32(rgb) static_cast<uint8_t>((((rgb) & kGreen6) >> 3) | (((rgb) & kGreen6) >> 9))
#define B16_32(rgb) static_cast<uint8_t>((((rgb) & kBlue5) >> 8) | (((rgb) & kBlue5) >> 14))
/* Extract red, green, and blue bytes from RGB32 dword. */
#define R32(rgb) static_cast<uint8_t>((rgb) & kRed8)
#define G32(rgb) static_cast<uint8_t>((((rgb) & kGreen8) >> 8) & 0xff)
#define B32(rgb) static_cast<uint8_t>((((rgb) & kBlue8) >> 16) & 0xff)
/* Build RGB565 word from red, green, and blue bytes. */
#define RGB565(r, g, b) static_cast<uint16_t>((((static_cast<uint16_t>(b) << 6) | (g)) << 5) | (r))
/* Build RGB32 dword from red, green, and blue bytes. */
#define RGB32(r, g, b) static_cast<uint32_t>((((static_cast<uint32_t>(b) << 8) | (g)) << 8) | (r))
#else // __BYTE_ORDER
/* Extract red, green, and blue bytes from RGB565 word. */
#define R16(rgb) static_cast<uint8_t>(((rgb) & kRed5) >> 11)
#define G16(rgb) static_cast<uint8_t>(((rgb) & kGreen6) >> 5)
#define B16(rgb) static_cast<uint8_t>((rgb) & kBlue5)
/* Make 8 bits red, green, and blue, extracted from RGB565 word. */
#define R16_32(rgb) static_cast<uint8_t>((((rgb) & kRed5) >> 8) | (((rgb) & kRed5) >> 14))
#define G16_32(rgb) static_cast<uint8_t>((((rgb) & kGreen6) >> 3) | (((rgb) & kGreen6) >> 9))
#define B16_32(rgb) static_cast<uint8_t>((((rgb) & kBlue5) << 3) | (((rgb) & kBlue5) >> 2))
/* Extract red, green, and blue bytes from RGB32 dword. */
#define R32(rgb) static_cast<uint8_t>(((rgb) & kRed8) >> 16)
#define G32(rgb) static_cast<uint8_t>(((rgb) & kGreen8) >> 8)
#define B32(rgb) static_cast<uint8_t>((rgb) & kBlue8)
/* Build RGB565 word from red, green, and blue bytes. */
#define RGB565(r, g, b) static_cast<uint16_t>((((static_cast<uint16_t>(r) << 6) | g) << 5) | b)
/* Build RGB32 dword from red, green, and blue bytes. */
#define RGB32(r, g, b) static_cast<uint32_t>((((static_cast<uint32_t>(r) << 8) | g) << 8) | b)
#endif // __BYTE_ORDER
/* An union that simplifies breaking 32 bit RGB into separate R, G, and B colors.
*/
typedef union RGB32_t {
uint32_t color;
struct {
#if __BYTE_ORDER == __LITTLE_ENDIAN
uint8_t r; uint8_t g; uint8_t b; uint8_t a;
#else // __BYTE_ORDER
uint8_t a; uint8_t b; uint8_t g; uint8_t r;
#endif // __BYTE_ORDER
};
} RGB32_t;
/* Clips a value to the unsigned 0-255 range, treating negative values as zero.
*/
static __inline__ int
clamp(int x)
{
if (x > 255) return 255;
if (x < 0) return 0;
return x;
}
/********************************************************************************
* Basics of RGB -> YUV conversion
*******************************************************************************/
/*
* RGB -> YUV conversion macros
*/
#define RGB2Y(r, g, b) (uint8_t)(((66 * (r) + 129 * (g) + 25 * (b) + 128) >> 8) + 16)
#define RGB2U(r, g, b) (uint8_t)(((-38 * (r) - 74 * (g) + 112 * (b) + 128) >> 8) + 128)
#define RGB2V(r, g, b) (uint8_t)(((112 * (r) - 94 * (g) - 18 * (b) + 128) >> 8) + 128)
/* Converts R8 G8 B8 color to YUV. */
static __inline__ void
R8G8B8ToYUV(uint8_t r, uint8_t g, uint8_t b, uint8_t* y, uint8_t* u, uint8_t* v)
{
*y = RGB2Y((int)r, (int)g, (int)b);
*u = RGB2U((int)r, (int)g, (int)b);
*v = RGB2V((int)r, (int)g, (int)b);
}
/* Converts RGB565 color to YUV. */
static __inline__ void
RGB565ToYUV(uint16_t rgb, uint8_t* y, uint8_t* u, uint8_t* v)
{
R8G8B8ToYUV(R16_32(rgb), G16_32(rgb), B16_32(rgb), y, u, v);
}
/* Converts RGB32 color to YUV. */
static __inline__ void
RGB32ToYUV(uint32_t rgb, uint8_t* y, uint8_t* u, uint8_t* v)
{
RGB32_t rgb_c;
rgb_c.color = rgb;
R8G8B8ToYUV(rgb_c.r, rgb_c.g, rgb_c.b, y, u, v);
}
/********************************************************************************
* Basics of YUV -> RGB conversion.
* Note that due to the fact that guest uses RGB only on preview window, and the
* RGB format that is used is RGB565, we can limit YUV -> RGB conversions to
* RGB565 only.
*******************************************************************************/
/*
* YUV -> RGB conversion macros
*/
/* "Optimized" macros that take specialy prepared Y, U, and V values:
* C = Y - 16
* D = U - 128
* E = V - 128
*/
#define YUV2RO(C, D, E) clamp((298 * (C) + 409 * (E) + 128) >> 8)
#define YUV2GO(C, D, E) clamp((298 * (C) - 100 * (D) - 208 * (E) + 128) >> 8)
#define YUV2BO(C, D, E) clamp((298 * (C) + 516 * (D) + 128) >> 8)
/*
* Main macros that take the original Y, U, and V values
*/
#define YUV2R(y, u, v) clamp((298 * ((y)-16) + 409 * ((v)-128) + 128) >> 8)
#define YUV2G(y, u, v) clamp((298 * ((y)-16) - 100 * ((u)-128) - 208 * ((v)-128) + 128) >> 8)
#define YUV2B(y, u, v) clamp((298 * ((y)-16) + 516 * ((u)-128) + 128) >> 8)
/* Converts YUV color to RGB565. */
static __inline__ uint16_t
YUVToRGB565(int y, int u, int v)
{
/* Calculate C, D, and E values for the optimized macro. */
y -= 16; u -= 128; v -= 128;
const uint16_t r = (YUV2RO(y,u,v) >> 3) & 0x1f;
const uint16_t g = (YUV2GO(y,u,v) >> 2) & 0x3f;
const uint16_t b = (YUV2BO(y,u,v) >> 3) & 0x1f;
return RGB565(r, g, b);
}
/* Converts YUV color to RGB32. */
static __inline__ uint32_t
YUVToRGB32(int y, int u, int v)
{
/* Calculate C, D, and E values for the optimized macro. */
y -= 16; u -= 128; v -= 128;
RGB32_t rgb;
rgb.r = YUV2RO(y,u,v) & 0xff;
rgb.g = YUV2GO(y,u,v) & 0xff;
rgb.b = YUV2BO(y,u,v) & 0xff;
return rgb.color;
}
/* YUV pixel descriptor. */
struct YUVPixel {
uint8_t Y;
uint8_t U;
uint8_t V;
inline YUVPixel()
: Y(0), U(0), V(0)
{
}
inline explicit YUVPixel(uint16_t rgb565)
{
RGB565ToYUV(rgb565, &Y, &U, &V);
}
inline explicit YUVPixel(uint32_t rgb32)
{
RGB32ToYUV(rgb32, &Y, &U, &V);
}
inline void get(uint8_t* pY, uint8_t* pU, uint8_t* pV) const
{
*pY = Y; *pU = U; *pV = V;
}
};
/* Converts an YV12 framebuffer to RGB565 framebuffer.
* Param:
* yv12 - YV12 framebuffer.
* rgb - RGB565 framebuffer.
* width, height - Dimensions for both framebuffers.
*/
void YV12ToRGB565(const void* yv12, void* rgb, int width, int height);
/* Converts an YV12 framebuffer to RGB32 framebuffer.
* Param:
* yv12 - YV12 framebuffer.
* rgb - RGB32 framebuffer.
* width, height - Dimensions for both framebuffers.
*/
void YV12ToRGB32(const void* yv12, void* rgb, int width, int height);
/* Converts an YU12 framebuffer to RGB32 framebuffer.
* Param:
* yu12 - YU12 framebuffer.
* rgb - RGB32 framebuffer.
* width, height - Dimensions for both framebuffers.
*/
void YU12ToRGB32(const void* yu12, void* rgb, int width, int height);
/* Converts an NV12 framebuffer to RGB565 framebuffer.
* Param:
* nv12 - NV12 framebuffer.
* rgb - RGB565 framebuffer.
* width, height - Dimensions for both framebuffers.
*/
void NV12ToRGB565(const void* nv12, void* rgb, int width, int height);
/* Converts an NV12 framebuffer to RGB32 framebuffer.
* Param:
* nv12 - NV12 framebuffer.
* rgb - RGB32 framebuffer.
* width, height - Dimensions for both framebuffers.
*/
void NV12ToRGB32(const void* nv12, void* rgb, int width, int height);
/* Converts an NV21 framebuffer to RGB565 framebuffer.
* Param:
* nv21 - NV21 framebuffer.
* rgb - RGB565 framebuffer.
* width, height - Dimensions for both framebuffers.
*/
void NV21ToRGB565(const void* nv21, void* rgb, int width, int height);
/* Converts an NV21 framebuffer to RGB32 framebuffer.
* Param:
* nv21 - NV21 framebuffer.
* rgb - RGB32 framebuffer.
* width, height - Dimensions for both framebuffers.
*/
void NV21ToRGB32(const void* nv21, void* rgb, int width, int height);
}; /* namespace chroma */
#endif /* HW_EMULATOR_CAMERA_CONVERTERS_H */
/*
* Copyright (C) 2011 The Android Open Source Project
*
* Licensed under the Apache License, Version 2.0 (the "License");
* you may not use this file except in compliance with the License.
* You may obtain a copy of the License at
*
* http://www.apache.org/licenses/LICENSE-2.0
*
* Unless required by applicable law or agreed to in writing, software
* distributed under the License is distributed on an "AS IS" BASIS,
* WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
* See the License for the specific language governing permissions and
* limitations under the License.
*/
/*
* Contains implemenation of framebuffer conversion routines.
*/
#define LOG_NDEBUG 0
#define LOG_TAG "EmulatedCamera_Converter"
#include "Converters.h"
namespace chroma {
static void _YUV420SToRGB565(const uint8_t* Y,
const uint8_t* U,
const uint8_t* V,
int dUV,
uint16_t* rgb,
int width,
int height)
{
const uint8_t* U_pos = U;
const uint8_t* V_pos = V;
for (int y = 0; y < height; y++) {
for (int x = 0; x < width; x += 2, U += dUV, V += dUV) {
const uint8_t nU = *U;
const uint8_t nV = *V;
*rgb = YUVToRGB565(*Y, nU, nV);
Y++; rgb++;
*rgb = YUVToRGB565(*Y, nU, nV);
Y++; rgb++;
}
if (y & 0x1) {
U_pos = U;
V_pos = V;
} else {
U = U_pos;
V = V_pos;
}
}
}
static void _YUV420SToRGB32(const uint8_t* Y,
const uint8_t* U,
const uint8_t* V,
int dUV,
uint32_t* rgb,
int width,
int height)
{
const uint8_t* U_pos = U;
const uint8_t* V_pos = V;
for (int y = 0; y < height; y++) {
for (int x = 0; x < width; x += 2, U += dUV, V += dUV) {
const uint8_t nU = *U;
const uint8_t nV = *V;
*rgb = YUVToRGB32(*Y, nU, nV);
Y++; rgb++;
*rgb = YUVToRGB32(*Y, nU, nV);
Y++; rgb++;
}
if (y & 0x1) {
U_pos = U;
V_pos = V;
} else {
U = U_pos;
V = V_pos;
}
}
}
void YV12ToRGB565(const void* yv12, void* rgb, int width, int height)
{
const int pix_total = width * height;
const uint8_t* Y = reinterpret_cast<const uint8_t*>(yv12);
const uint8_t* U = Y + pix_total;
const uint8_t* V = U + pix_total / 4;
_YUV420SToRGB565(Y, U, V, 1, reinterpret_cast<uint16_t*>(rgb), width, height);
}
void YV12ToRGB32(const void* yv12, void* rgb, int width, int height)
{
const int pix_total = width * height;
const uint8_t* Y = reinterpret_cast<const uint8_t*>(yv12);
const uint8_t* V = Y + pix_total;
const uint8_t* U = V + pix_total / 4;
_YUV420SToRGB32(Y, U, V, 1, reinterpret_cast<uint32_t*>(rgb), width, height);
}
void YU12ToRGB32(const void* yu12, void* rgb, int width, int height)
{
const int pix_total = width * height;
const uint8_t* Y = reinterpret_cast<const uint8_t*>(yu12);
const uint8_t* U = Y + pix_total;
const uint8_t* V = U + pix_total / 4;
_YUV420SToRGB32(Y, U, V, 1, reinterpret_cast<uint32_t*>(rgb), width, height);
}
/* Common converter for YUV 4:2:0 interleaved to RGB565.
* y, u, and v point to Y,U, and V panes, where U and V values are interleaved.
*/
static void _NVXXToRGB565(const uint8_t* Y,
const uint8_t* U,
const uint8_t* V,
uint16_t* rgb,
int width,
int height)
{
_YUV420SToRGB565(Y, U, V, 2, rgb, width, height);
}
/* Common converter for YUV 4:2:0 interleaved to RGB32.
* y, u, and v point to Y,U, and V panes, where U and V values are interleaved.
*/
static void _NVXXToRGB32(const uint8_t* Y,
const uint8_t* U,
const uint8_t* V,
uint32_t* rgb,
int width,
int height)
{
_YUV420SToRGB32(Y, U, V, 2, rgb, width, height);
}
void NV12ToRGB565(const void* nv12, void* rgb, int width, int height)
{
const int pix_total = width * height;
const uint8_t* y = reinterpret_cast<const uint8_t*>(nv12);
_NVXXToRGB565(y, y + pix_total, y + pix_total + 1,
reinterpret_cast<uint16_t*>(rgb), width, height);
}
void NV12ToRGB32(const void* nv12, void* rgb, int width, int height)
{
const int pix_total = width * height;
const uint8_t* y = reinterpret_cast<const uint8_t*>(nv12);
_NVXXToRGB32(y, y + pix_total, y + pix_total + 1,
reinterpret_cast<uint32_t*>(rgb), width, height);
}
void NV21ToRGB565(const void* nv21, void* rgb, int width, int height)
{
const int pix_total = width * height;
const uint8_t* y = reinterpret_cast<const uint8_t*>(nv21);
_NVXXToRGB565(y, y + pix_total + 1, y + pix_total,
reinterpret_cast<uint16_t*>(rgb), width, height);
}
void NV21ToRGB32(const void* nv21, void* rgb, int width, int height)
{
const int pix_total = width * height;
const uint8_t* y = reinterpret_cast<const uint8_t*>(nv21);
_NVXXToRGB32(y, y + pix_total + 1, y + pix_total,
reinterpret_cast<uint32_t*>(rgb), width, height);
}
}; /* namespace chroma */
一旦你夹紧，你就完成了。它们变成不同的颜色，你不能回去。如果您愿意，我已经编写了一些我自己的代码来在所有这些和更多之间进行转换，但这无助于将钳位颜色反转回其原始颜色。