How to read a bitmap from the Windows Clipboard - c++

I am writing an extremely small C++ program to help me animate sprites. I'd like it to take data I copy to the clipboard from photoshop, manipulate it in my program, then overwrite the clipboard with the transform.
The problem though is that I'm not sure how to read the initial clipboard from photoshop.
I can load the clipboard with GetClipboardData(CF_DIB), and get a valid handle, but I've no idea how to use that handle. I've tried using SFML's Image::LoadFromMemory(handle, GlobalSize(handle)) which is able to load bitmap files from memory, but that doesn't seem to work.
Will I need to actually parse the entire format? What format structure would I be looking at in that case? Would there perhaps be any way I could quickly mangle the data so it might look like a bitmap file? Could it be easier/possible to simply save it to file using the windows API? (I could then load that file with SFML to edit, that way)
It's just a quick and dirty tool for myself to save a lot of grunt work in photoshop, so efficiency or robustness aren't important at all.

Learn the bitmap structure from Wikipedia and then write it out to a file and then write out the pixels..
I've tested the below with Paint on Windows 8.1. I opened an image with paint and then pressed Ctrl + C to copy to the clipboard.. then I ran the following code and it copied the clipboard image to the desktop:
#include <iostream>
#include <fstream>
#include <windows.h>
int main()
{
std::cout<<"Format Bitmap: "<<IsClipboardFormatAvailable(CF_BITMAP)<<"\n";
std::cout<<"Format DIB: "<<IsClipboardFormatAvailable(CF_DIB)<<"\n";
std::cout<<"Format DIBv5: "<<IsClipboardFormatAvailable(CF_DIBV5)<<"\n";
if (IsClipboardFormatAvailable(CF_DIB))
{
if (OpenClipboard(NULL))
{
HANDLE hClipboard = GetClipboardData(CF_DIB);
if (hClipboard != NULL && hClipboard != INVALID_HANDLE_VALUE)
{
void* dib = GlobalLock(hClipboard);
if (dib)
{
BITMAPINFOHEADER* info = reinterpret_cast<BITMAPINFOHEADER*>(dib);
BITMAPFILEHEADER fileHeader = {0};
fileHeader.bfType = 0x4D42;
fileHeader.bfOffBits = 54;
fileHeader.bfSize = (((info->bmiHeader.biWidth * info->bmiHeader.biBitCount + 31) & ~31) / 8
* info->bmiHeader.biHeight) + fileHeader.bfOffBits;
std::cout<<"Type: "<<std::hex<<fileHeader.bfType<<std::dec<<"\n";
std::cout<<"bfSize: "<<fileHeader.bfSize<<"\n";
std::cout<<"Reserved: "<<fileHeader.bfReserved1<<"\n";
std::cout<<"Reserved2: "<<fileHeader.bfReserved2<<"\n";
std::cout<<"Offset: "<<fileHeader.bfOffBits<<"\n";
std::cout<<"biSize: "<<info->bmiHeader.biSize<<"\n";
std::cout<<"Width: "<<info->bmiHeader.biWidth<<"\n";
std::cout<<"Height: "<<info->bmiHeader.biHeight<<"\n";
std::cout<<"Planes: "<<info->bmiHeader.biPlanes<<"\n";
std::cout<<"Bits: "<<info->bmiHeader.biBitCount<<"\n";
std::cout<<"Compression: "<<info->bmiHeader.biCompression<<"\n";
std::cout<<"Size: "<<info->bmiHeader.biSizeImage<<"\n";
std::cout<<"X-res: "<<info->bmiHeader.biXPelsPerMeter<<"\n";
std::cout<<"Y-res: "<<info->bmiHeader.biYPelsPerMeter<<"\n";
std::cout<<"ClrUsed: "<<info->bmiHeader.biClrUsed<<"\n";
std::cout<<"ClrImportant: "<<info->bmiHeader.biClrImportant<<"\n";
std::ofstream file("C:/Users/Brandon/Desktop/Test.bmp", std::ios::out | std::ios::binary);
if (file)
{
file.write(reinterpret_cast<char*>(&fileHeader), sizeof(BITMAPFILEHEADER));
file.write(reinterpret_cast<char*>(info), sizeof(BITMAPINFOHEADER));
file.write(reinterpret_cast<char*>(++info), bmp.dib.biSizeImage);
}
GlobalUnlock(dib);
}
}
CloseClipboard();
}
}
return 0;
}

I wasn't going to post an answer initially, after all you already have a good enough answer. But I guess I was coerced into doing so anyway, besides, this is the primary question that pops when you search for GetClipboardData CF_DIB, so might as well try to present a more complete solution.
Unfortunately, clipboard formats are a minefield. And GDI bitmaps are an even bigger minefield. CF_DIB gives you a "packed DIB", you do need to parse it to some extent if you actually want to do anything meaningful with it. This is the layout (pseudo code):
struct PACKED_DIB {
struct BITMAPINFO {
BITMAPINFOHEADER bih; // 40 bytes
DWORD optional_RGB_bitmaks[]; // (variable size)
DWORD optional_color_table[]; // (variable size)
}
BYTE pixel_data_array[]; // (variable size)
}
The total size of the structure is given by GlobalSize(). The crucial bit of information that is required for any further processing is the offset, in bytes, from the start of the BITMAPINFO structure to the start of the pixel data array. If the optional bitmasks and the color table are absent, this offset is constant 40 (sizeof(BITMAPINFOHEADER)). Whether this is the case depends entirely on how an application has put the bitmap into the clipboard. Most applications do this because it's the simplest way.
This code calculates that offset:
// Returns the offset, in bytes, from the start of the BITMAPINFO, to the start of the pixel data array, for a packed DIB.
static INT GetPixelDataOffsetForPackedDIB(const BITMAPINFOHEADER *BitmapInfoHeader)
{
INT OffsetExtra = 0;
if (BitmapInfoHeader->biSize == sizeof(BITMAPINFOHEADER) /* 40 */)
{
// This is the common BITMAPINFOHEADER type. In this case, there may be bit masks following the BITMAPINFOHEADER
// and before the actual pixel bits (does not apply if bitmap has <= 8 bpp)
if (BitmapInfoHeader->biBitCount > 8)
{
if (BitmapInfoHeader->biCompression == BI_BITFIELDS)
{
OffsetExtra += 3 * sizeof(RGBQUAD);
}
else if (BitmapInfoHeader->biCompression == 6 /* BI_ALPHABITFIELDS */)
{
// Not widely supported, but technically a valid DIB format.
// You *can* get this in the clipboard, although neither GDI nor stb_image will like it.
OffsetExtra += 4 * sizeof(RGBQUAD);
}
}
}
if (BitmapInfoHeader->biClrUsed > 0)
{
// We have no choice but to trust this value.
OffsetExtra += BitmapInfoHeader->biClrUsed * sizeof(RGBQUAD);
}
else
{
// In this case, the color table contains the maximum number for the current bit count (0 if > 8bpp)
if (BitmapInfoHeader->biBitCount <= 8)
{
// 1bpp: 2
// 4bpp: 16
// 8bpp: 256
OffsetExtra += sizeof(RGBQUAD) << BitmapInfoHeader->biBitCount;
}
}
return BitmapInfoHeader->biSize + OffsetExtra;
}
Below is a program that demonstrates several things you can do with this offset:
Write the clipboard image to a .bmp file
Load it from memory using SFML (sf::Image::loadFromMemory)
Put it back in the clipboard from an SFML image
Convert it to a HBITMAP (so it can be used in GDI)
Put it back in the clipboard from a GDI HBITMAP
#include <sdkddkver.h>
#define WIN32_LEAN_AND_MEAN
#define NOMINMAX
#include <Windows.h>
#include <stdio.h>
#include <stdlib.h>
#include <assert.h>
#if DEMO_SFML
#include <SFML/Graphics.hpp>
#endif
static BOOL OpenClipboard_ButTryABitHarder(HWND ClipboardOwner);
static INT GetPixelDataOffsetForPackedDIB(const BITMAPINFOHEADER *BitmapInfoHeader);
static void PutBitmapInClipboard_AsDIB(HBITMAP hBitmap);
static void PutBitmapInClipboard_From32bppTopDownRGBAData(INT Width, INT Height, const void *Data32bppRGBA);
int wmain(int argc, wchar_t *argv[])
{
if (!OpenClipboard_ButTryABitHarder(NULL))
{
// Could not open clipboard. This usually indicates that another application is permanently blocking it.
return 1;
}
HGLOBAL ClipboardDataHandle = (HGLOBAL)GetClipboardData(CF_DIB);
if (!ClipboardDataHandle)
{
// Clipboard object is not a DIB, and is not auto-convertible to DIB
CloseClipboard();
return 0;
}
BITMAPINFOHEADER *BitmapInfoHeader = (BITMAPINFOHEADER *)GlobalLock(ClipboardDataHandle);
assert(BitmapInfoHeader); // This can theoretically fail if mapping the HGLOBAL into local address space fails. Very pathological, just act as if it wasn't a bitmap in the clipboard.
SIZE_T ClipboardDataSize = GlobalSize(ClipboardDataHandle);
assert(ClipboardDataSize >= sizeof(BITMAPINFOHEADER)); // Malformed data. While older DIB formats exist (e.g. BITMAPCOREHEADER), they are not valid data for CF_DIB; it mandates a BITMAPINFO struct. If this fails, just act as if it wasn't a bitmap in the clipboard.
INT PixelDataOffset = GetPixelDataOffsetForPackedDIB(BitmapInfoHeader);
// ============================================================================================================
// ============================================================================================================
//
// Example 1: Write it to a .bmp file
//
// The clipboard contains a packed DIB, whose start address coincides with BitmapInfoHeader, and whose total size is ClipboardDataSize.
// By definition, we can jam the whole DIB memory into a BMP file as-is, except that we need to prepend a BITMAPFILEHEADER struct.
// The tricky part is that for BITMAPFILEHEADER.bfOffBits, which must be calculated using the information in BITMAPINFOHEADER.
// The BMP file layout:
// #offset 0: BITMAPFILEHEADER
// #offset 14 (sizeof(BITMAPFILEHEADER)): BITMAPINFOHEADER
// #offset 14 + BitmapInfoHeader->biSize: Optional bit masks and color table
// #offset 14 + DIBPixelDataOffset: pixel bits
// #offset 14 + ClipboardDataSize: EOF
size_t TotalBitmapFileSize = sizeof(BITMAPFILEHEADER) + ClipboardDataSize;
wprintf(L"BITMAPINFOHEADER size: %u\r\n", BitmapInfoHeader->biSize);
wprintf(L"Format: %hubpp, Compression %u\r\n", BitmapInfoHeader->biBitCount, BitmapInfoHeader->biCompression);
wprintf(L"Pixel data offset within DIB: %u\r\n", PixelDataOffset);
wprintf(L"Total DIB size: %zu\r\n", ClipboardDataSize);
wprintf(L"Total bitmap file size: %zu\r\n", TotalBitmapFileSize);
BITMAPFILEHEADER BitmapFileHeader = {};
BitmapFileHeader.bfType = 0x4D42;
BitmapFileHeader.bfSize = (DWORD)TotalBitmapFileSize; // Will fail if bitmap size is nonstandard >4GB
BitmapFileHeader.bfOffBits = sizeof(BITMAPFILEHEADER) + PixelDataOffset;
HANDLE FileHandle = CreateFileW(L"test.bmp", GENERIC_WRITE, 0, NULL, CREATE_ALWAYS, FILE_ATTRIBUTE_NORMAL, NULL);
if (FileHandle != INVALID_HANDLE_VALUE)
{
DWORD dummy = 0;
BOOL Success = true;
Success &= WriteFile(FileHandle, &BitmapFileHeader, sizeof(BITMAPFILEHEADER), &dummy, NULL);
Success &= WriteFile(FileHandle, BitmapInfoHeader, (DWORD)ClipboardDataSize, &dummy, NULL);
Success &= CloseHandle(FileHandle);
if (Success)
{
wprintf(L"File saved.\r\n");
}
}
#if DEMO_SFML
// ============================================================================================================
// ============================================================================================================
//
// Example 2: Load it from memory in SFML
//
// SFML expects a whole bitmap file, including its BITMAPFILEHEADER, in memory.
// So this is similar to Example 1, except in memory.
BYTE *BitmapFileContents = (BYTE *)malloc(TotalBitmapFileSize);
assert(BitmapFileContents);
memcpy(BitmapFileContents, &BitmapFileHeader, sizeof(BITMAPFILEHEADER));
// Append DIB
memcpy(BitmapFileContents + sizeof(BITMAPFILEHEADER), BitmapInfoHeader, ClipboardDataSize);
sf::Image image;
image.loadFromMemory(BitmapFileContents, TotalBitmapFileSize);
// The memory can be freed once the image has been loaded in SFML.
free(BitmapFileContents);
// Manipulate it:
image.flipHorizontally();
// Put it back in the clipboard:
PutBitmapInClipboard_From32bppTopDownRGBAData(image.getSize().x, image.getSize().y, image.getPixelsPtr());
#else
// ============================================================================================================
// ============================================================================================================
//
// Example 3: Convert to HBITMAP for GDI
//
BYTE *PixelDataFromClipboard = (BYTE *)BitmapInfoHeader + PixelDataOffset;
// This will only work if the DIB format is supported by GDI. Not all formats are supported.
BYTE *PixelDataNew;
HBITMAP hBitmap = CreateDIBSection(NULL, (BITMAPINFO *)BitmapInfoHeader, DIB_RGB_COLORS, (void **)&PixelDataNew, NULL, 0);
assert(hBitmap);
// Need to copy the data from the clipboard to the new DIBSection.
BITMAP BitmapDesc = {};
GetObjectW(hBitmap, sizeof(BitmapDesc), &BitmapDesc);
SIZE_T PixelDataBytesToCopy = (SIZE_T)BitmapDesc.bmHeight * BitmapDesc.bmWidthBytes;
SIZE_T PixelDataBytesAvailable = ClipboardDataSize - PixelDataOffset;
if (PixelDataBytesAvailable < PixelDataBytesToCopy)
{
// Malformed data; doesn't contain enough pixels.
PixelDataBytesToCopy = PixelDataBytesAvailable;
}
memcpy(PixelDataNew, PixelDataFromClipboard, PixelDataBytesToCopy);
// NOTE: While it is possible to create a DIB section without copying the pixel data, in general you'd want to
// copy it anyway because the clipboard needs to be closed asap.
// Draw something on it.
PixelDataNew[7] = 0;
PixelDataNew[11] = 100;
HDC hdc = CreateCompatibleDC(NULL);
assert(hdc);
SelectObject(hdc, hBitmap);
RECT rc = { 0, 0, BitmapDesc.bmWidth / 2, BitmapDesc.bmHeight / 2 };
HBRUSH brush = CreateSolidBrush(RGB(250, 100, 0));
FillRect(hdc, &rc, brush);
DeleteObject(brush);
DeleteDC(hdc);
// ============================================================================================================
// ============================================================================================================
//
// Copy it back to the clipboard.
//
PutBitmapInClipboard_AsDIB(hBitmap);
#endif // DEMO_SFML
GlobalUnlock(ClipboardDataHandle);
CloseClipboard();
return 0;
}
static BOOL OpenClipboard_ButTryABitHarder(HWND hWnd)
{
for (int i = 0; i < 20; ++i)
{
// This can fail if the clipboard is currently being accessed by another application.
if (OpenClipboard(hWnd)) return true;
Sleep(10);
}
return false;
}
// Returns the offset, in bytes, from the start of the BITMAPINFO, to the start of the pixel data array, for a packed DIB.
static INT GetPixelDataOffsetForPackedDIB(const BITMAPINFOHEADER *BitmapInfoHeader)
{
INT OffsetExtra = 0;
if (BitmapInfoHeader->biSize == sizeof(BITMAPINFOHEADER) /* 40 */)
{
// This is the common BITMAPINFOHEADER type. In this case, there may be bit masks following the BITMAPINFOHEADER
// and before the actual pixel bits (does not apply if bitmap has <= 8 bpp)
if (BitmapInfoHeader->biBitCount > 8)
{
if (BitmapInfoHeader->biCompression == BI_BITFIELDS)
{
OffsetExtra += 3 * sizeof(RGBQUAD);
}
else if (BitmapInfoHeader->biCompression == 6 /* BI_ALPHABITFIELDS */)
{
// Not widely supported, but valid.
OffsetExtra += 4 * sizeof(RGBQUAD);
}
}
}
if (BitmapInfoHeader->biClrUsed > 0)
{
// We have no choice but to trust this value.
OffsetExtra += BitmapInfoHeader->biClrUsed * sizeof(RGBQUAD);
}
else
{
// In this case, the color table contains the maximum number for the current bit count (0 if > 8bpp)
if (BitmapInfoHeader->biBitCount <= 8)
{
// 1bpp: 2
// 4bpp: 16
// 8bpp: 256
OffsetExtra += sizeof(RGBQUAD) << BitmapInfoHeader->biBitCount;
}
}
return BitmapInfoHeader->biSize + OffsetExtra;
}
// Helper function for interaction with libraries like stb_image.
// Data will be copied, so you can do what you want with it after this function returns.
static void PutBitmapInClipboard_From32bppTopDownRGBAData(INT Width, INT Height, const void *Data32bppRGBA)
{
// Nomenclature: Data at offset 0 is R top left corner, offset 1 is G top left corner, etc.
// This is pretty much the opposite of what a HBITMAP normally does.
assert(Width > 0);
assert(Height > 0);
assert(Data32bppRGBA);
// GDI won't help us here if we want to preserve the alpha channel. It doesn't support BI_ALPHABITFIELDS, and
// we can't use BI_RGB directly because BI_RGB actually means BGRA in reality.
// That means, unfortunately it's not going to be a simple memcpy :(
DWORD PixelDataSize = 4/*32bpp*/ * Width * Height;
// We need BI_BITFIELDS for RGB color masks here.
size_t TotalSize = sizeof(BITMAPINFOHEADER) + PixelDataSize;
HGLOBAL hGlobal = GlobalAlloc(GMEM_MOVEABLE | GMEM_ZEROINIT, TotalSize);
assert(hGlobal);
void *mem = GlobalLock(hGlobal);
assert(mem);
BITMAPINFOHEADER *bih = (BITMAPINFOHEADER *)mem;
bih->biSize = sizeof(BITMAPINFOHEADER);
bih->biWidth = Width;
bih->biHeight = -Height; // Negative height means top-down bitmap
bih->biPlanes = 1;
bih->biBitCount = 32;
bih->biCompression = BI_RGB;
bih->biSizeImage = PixelDataSize;
BYTE *PixelData = (BYTE *)mem + sizeof(BITMAPINFOHEADER);
DWORD NumPixels = Width * Height;
for (DWORD i = 0; i < NumPixels; ++i)
{
// Convert RGBA to BGRA
DWORD tmp = ((DWORD *)Data32bppRGBA)[i];
DWORD tmp2 = tmp & 0xff00ff00; // assumes LE
tmp2 |= (tmp >> 16) & 0xff;
tmp2 |= (tmp & 0xff) << 16;
((DWORD *)PixelData)[i] = tmp2;
}
GlobalUnlock(hGlobal);
EmptyClipboard();
SetClipboardData(CF_DIB, hGlobal);
// The hGlobal now belongs to the clipboard. Do not free it.
}
// Bitmap will be copied, so you can do what you want with it after this function returns.
static void PutBitmapInClipboard_AsDIB(HBITMAP hBitmap)
{
// Need this to get the bitmap dimensions.
BITMAP desc = {};
int tmp = GetObjectW(hBitmap, sizeof(desc), &desc);
assert(tmp != 0);
// We need to build this structure in a GMEM_MOVEABLE global memory block:
// BITMAPINFOHEADER (40 bytes)
// PixelData (4 * Width * Height bytes)
// We're enforcing 32bpp BI_RGB, so no bitmasks and no color table.
// NOTE: SetClipboardData(CF_DIB) insists on the size 40 version of BITMAPINFOHEADER, otherwise it will misinterpret the data.
DWORD PixelDataSize = 4/*32bpp*/ * desc.bmWidth * desc.bmHeight; // Correct alignment happens implicitly.
assert(desc.bmWidth > 0);
assert(desc.bmHeight > 0);
size_t TotalSize = sizeof(BITMAPINFOHEADER) + PixelDataSize;
HGLOBAL hGlobal = GlobalAlloc(GMEM_MOVEABLE | GMEM_ZEROINIT, TotalSize);
assert(hGlobal);
void *mem = GlobalLock(hGlobal);
assert(mem);
BITMAPINFOHEADER *bih = (BITMAPINFOHEADER *)mem;
bih->biSize = sizeof(BITMAPINFOHEADER);
bih->biWidth = desc.bmWidth;
bih->biHeight = desc.bmHeight;
bih->biPlanes = 1;
bih->biBitCount = 32;
bih->biCompression = BI_RGB;
bih->biSizeImage = PixelDataSize;
HDC hdc = CreateCompatibleDC(NULL);
assert(hdc);
HGDIOBJ old = SelectObject(hdc, hBitmap);
assert(old != nullptr); // This can fail if the hBitmap is still selected into a different DC.
void *PixelData = (BYTE *)mem + sizeof(BITMAPINFOHEADER);
// Pathologial "bug": If the bitmap is a DDB that originally belonged to a device with a different palette, that palette is lost. The caller would need to give us the correct HDC, but this is already insane enough as it is.
tmp = GetDIBits(hdc, hBitmap, 0, desc.bmHeight, PixelData, (BITMAPINFO *)bih, DIB_RGB_COLORS);
assert(tmp != 0);
// NOTE: This will correctly preserve the alpha channel if possible, but it's up to the receiving application to handle it.
DeleteDC(hdc);
GlobalUnlock(hGlobal);
EmptyClipboard();
SetClipboardData(CF_DIB, hGlobal);
// The hGlobal now belongs to the clipboard. Do not free it.
}
I intend this code to be mostly production-ready because I need it for myself, if anyone finds a problem I'd be happy to hear about it.
Some additional notes for reference:
Tested on Win10
Tested on WinXP (except SFML), although the %zu doesn't work in older CRTs, who knew
Error handling is not production-ready.
What the explanation of CF_DIB really wanted to say is "it's a packed DIB". There is no official guarantee that it will be a plain BITMAPINFOHEADER, i.e. biSize == 40 though, although it is likely that this is the case.
The BITMAPFINO documentation explains that the structure is really variable in length, and that BITMAPINFOHEADER.biClrUsed needs to be taken into account, but it fails to mention BI_BITFIELDS.
BITMAPINFOHEADER has more details on this, but fails to mention BI_ALPHABITFIELDS or the fact that the bitmasks are only present if the polymorphic BITMAPINFOHEADER struct is actually a plain BITMAPINFOHEADER (i.e. biSize == 40). Later versions, like BITMAPV5HEADER, include the bitmasks unconditionally.
All in all, the wikipedia article on the BMP file format contains a much more concise and coherent explanation of the DIB memory layout.
Older versions of Paint handled the clipboard, including the offset calculation in a very similar fashion to what I did above in GetDIBPixelDataOffset (obviously I can't post that verbatim here). It does not assume that biSize == 40. Newer versions of Paint use COleServerItem for clipboard handling.
As a final reference, the source code of GTK, which is used by GIMP and other cross-platform software, implements CF_DIB very similarly. That code even handles web browser specific formats, so a bit harder to follow. It's the transmute_cf_dib_to_image_bmp function. The function's length parameter comes from GlobalSize. Note that it also does not assume that biSize == 40.
If SetClipboardData is called with CF_BITMAP, it requires a DDB (it will silently fail if you pass it a HBITMAP that is really a DIB). A CF_BITMAP that is implicitly converted to a DIB uses BI_BITFIELDS, this also applies to screenshots (at least if the original DDB was compatible with the screen DC).
Putting bitmaps into the clipboard is a whole new can of worms. If a DDB is put in the clipboard with CF_BITMAP, that bitmap is not copied (at least not initially). If any program manipulates it, all programs accessing the clipboard will see the manipulated bitmap. However, as soon as any one application ever requests it as a CF_DIB, Windows applies a bunch of magic, and that is no longer true, the bitmap is now a copy. This does not apply to bitmaps that were put in the clipboard as CF_DIB, those immediately become immune to manipulations by other programs. CF_DIB seems to have fewer unpleasant implications and surprises, and also seems to be used by most applications. While you could try to preserve the original bitmap's format when putting it in the clipboard, I chose to use a fixed catch-all format for outgoing data because it's already crazy enough.
SetClipboardData implies that CF_DIB doesn't work with Windows Store apps, but I was unable to confirm that claim. Then again, the paragraph about a NULL owner is also incorrect.
Both PutBitmapInClipboard_AsDIB and PutBitmapInClipboard_From32bppTopDownRGBAData do preserve the alpha channel if possible, although GDI drawing functions as demonstrated don't support alpha and will destroy it (SFML will handle it just fine). Putting the alpha channel in the MSB and using BI_RGB seems to be the de-facto standard for storing alpha in DIBs.

Related

WinAPI/GDI: How to use GetDIBits() to get color table synthesized for a bitmap?

I find it difficult to understand the excerpt below from MSDN site on GetDIBits() function:
If lpvBits is NULL and the bit count member of BITMAPINFO is
initialized to zero, GetDIBits fills in a BITMAPINFOHEADER
structure or BITMAPCOREHEADER without the color table. This
technique can be used to query bitmap attributes.
Question-1: What is meant by "the bit count member of BITMAPINFO"? Does it mean some_bmi.bmiHeader.biBitCount?
Question-2: What is meant by "GetDIBits fills in a BITMAPINFOHEADER structure or BITMAPCOREHEADER without the color table"? What color table is there to fill in those structures? None of them seems to have a member related to the color table. Is this about the array some_bmi.bmiColors?
Question-3: Is there a way to use GetDIBits() to get the color table(i.e. the array mapping indexes to colors) for a bitmap?
EDIT:
From the comments so far, it looks like breaking the question down into smaller parts was not effective. I will try it another way.
This is what I understand from the part I quoted from MSDN at the beginning:
Assuming the function call is GetDIBits(hdc, hbmp, uStartScan, cScanLines, lpvBits, lpbi, uUsage); if lpvBits is NULL and lpvBits->bmiHeader.biBitCount is initialized to zero, GetDIBits() fills in lpbi->bmiHeader only and lpbi->bmiColors is not modified.
Is this the correct way to understand it? And if so, is there a way to get GetDIBits() to fill in lpbi->bmiColors, such as initializing lpvBits->bmiHeader.biBitCount to bit-depth of the bitmap?
I tried testing Question-1's assumption as follows but GetDIBits() fails in that case:
void test(HWND hWnd) {
// get a memory DC
HDC hdc = GetDC(hWnd);
HDC hdcmem = CreateCompatibleDC(hdc); // has 1x1 mono bitmap selected
// into it initially by default
// select a 16x16 mono bmp into it
const int bmp_h = 16, bmp_w = 16;
const int bits_per_px = 1;
HBITMAP hbmp = CreateCompatibleBitmap(hdcmem, bmp_h, bmp_w); // 16x16 mono bitmap
HGDIOBJ hOldBmp = SelectObject(hdcmem, hbmp);
// initialize BITMAPINFO ptr
// (make sure to allocate a buffer large enough for 2 RGBQUADs
// in case color table is retured by GetDIBits() call)
const int bmi_buf_sz =
sizeof(BITMAPINFO) + sizeof(RGBQUAD) * (1 << bits_per_px); // 2 + 1(extra) RGBQUADs allocated for pbmi->bimColors
BYTE* p_bmi_buf = new BYTE[bmi_buf_sz];
BITMAPINFO* pbmi = reinterpret_cast<BITMAPINFO*>(p_bmi_buf);
ZeroMemory(pbmi, bmi_buf_sz);
// populate BITMAPINFO
pbmi->bmiHeader.biSize = sizeof(BITMAPINFOHEADER);
pbmi->bmiHeader.biBitCount = 1; // set to 1 just to see if GetDIBits()
// fills in pbmi->bmiColors too
// (when set to 0, only pbmi->bmiHeader is filled)
if(!GetDIBits(hdcmem, hbmp,
0, (UINT)bmp_h,
NULL, pbmi, DIB_PAL_COLORS)) {
MessageBox(hWnd, L"GetDIBits() failed!", NULL, MB_OK);
}
// clean-up
delete[] p_bmi_buf;
SelectObject(hdcmem, hOldBmp); // push hbmp out
DeleteObject(hbmp);
DeleteDC(hdcmem);
ReleaseDC(hWnd, hdc);
}
The easiest way to get the color table is with GetDibColorTable:
HDC memdc = CreateCompatibleDC(NULL);
HBITMAP oldbmp = (HBITMAP)SelectObject(memdc, hbitmap);
int ncolors = 1 << bm.bmBitsPixel;
std::vector<RGBQUAD> rgb(ncolors);
if(ncolors == GetDIBColorTable(memdc, 0, ncolors, &rgb[0]))
{
//success!
}
SelectObject(memdc, oldbmp);
DeleteDC(memdc);
Back to your question: GetDIBits expects pbmi to contain all zeros (except for bmiHeader.biSize member). So pbmi->bmiHeader.biBitCount should be zero.
//pbmi->bmiHeader.biBitCount = 1; <<= comment out this line
This should work; however, as it is stated in documentation, this will only fill the info header, not the color table. To get the color table you have to make another call to GetDIBits with enough allocation for the dib bits.
Moreover DIB_PAL_COLORS will return palette index array (I am not sure what you can do with that). You may want to use DIB_RGB_COLORS flag which will return actual colors when a color table is present.
Try this with a bitmap loaded from file:
HBITMAP hbitmap = (HBITMAP)LoadImage(0, L"8bit.bmp",
IMAGE_BITMAP, 0, 0, LR_CREATEDIBSECTION | LR_LOADFROMFILE);
BITMAP bm;
GetObject(hbitmap, sizeof(bm), &bm);
//don't continue for hi color bitmaps
if(bm.bmBitsPixel > 8) return;
int ncolors = 1 << bm.bmBitsPixel;
HDC memdc = CreateCompatibleDC(NULL);
int bmpinfo_size = sizeof(BITMAPINFOHEADER) + sizeof(RGBQUAD) * ncolors;
std::vector<BYTE> buf(bmpinfo_size);
BITMAPINFO* bmpinfo = (BITMAPINFO*)buf.data();
bmpinfo->bmiHeader.biSize = sizeof(BITMAPINFOHEADER);
if(!GetDIBits(memdc, hbitmap, 0, bm.bmHeight, NULL, bmpinfo, DIB_RGB_COLORS))
{
DWORD err = GetLastError();
//...
}
int dibsize = ((bm.bmWidth * bm.bmBitsPixel + 31) / 32) * 4 * bm.bmHeight;
std::vector<BYTE> dib(dibsize);
if(!GetDIBits(memdc, hbitmap, 0, (UINT)bm.bmHeight, &dib[0], bmpinfo, DIB_RGB_COLORS))
{
DWORD err = GetLastError();
//...
}
Now bmpinfo->bmiColors should contain the same values as rgb array shown earlier.
Possible confusion between BITMAPINFO and BITMAPINFOHEADER:
The above structures are declared as follows:
typedef struct tagBITMAPINFO {
BITMAPINFOHEADER bmiHeader;
RGBQUAD bmiColors[1];
} BITMAPINFO, FAR *LPBITMAPINFO, *PBITMAPINFO;
typedef struct tagBITMAPINFOHEADER{
DWORD biSize;
LONG biWidth;
LONG biHeight;
WORD biPlanes;
WORD biBitCount;
DWORD biCompression;
DWORD biSizeImage;
LONG biXPelsPerMeter;
LONG biYPelsPerMeter;
DWORD biClrUsed;
DWORD biClrImportant;
} BITMAPINFOHEADER, FAR *LPBITMAPINFOHEADER, *PBITMAPINFOHEADER;
So BITMAPINFO does not have biBitCount member. But it does have bmiHeader.biBitCount member.
When you declare a BITMAPINFOHEADER variable, you have to set biSize member (that's Windows' idea of version control). When you declare a BITMAPINFO variable, you have to make sure it's BITMAPINFOHEADER is taken care of.
Note that most of the time you don't have to worry about palette. For example LoadImage will return a compatible bitmap (if you don't specify LR_CREATEDIBSECTION) and you can use that right away.
Although the accepted answer covers the details, this is more of a direct answer to the questions in the OP.
Assuming the function call is GetDIBits(hdc, hbmp, uStartScan, cScanLines, lpvBits, lpbi, uUsage);
Question-1:
That is correct. "The bit count member of BITMAPINFO" refers to lpbi->bmiHeader.biBitCount.
Question-2:
When GetDIBits() is called to get DIB bits(i.e. with a non-null lpvBits and an appropriately initialized lpbi->bmiHeader), lpbi->bmiColors also gets filled by the function with the color table(if bit depth is less then 24 bpp).
Unfortunately, this is not clear in the documentation of the function. With that in mind, what the quoted part means is that, when lpvBits is NULL and lpbi->bmiHeader.biBitCount is zero, the function fills lpbi->bmiHeader only, and does not modify lpbi->bimColor(as opposed to when caling the function to get DIB bits).
Question-3:
You can get the function to return color table(for bitmaps with 8-bbp or less) in lpbi->bmiColors by calling it with a non-null lpvBits and an appropriately initialized lpbi->bmiHeader. IOW, when you call the function to get DIB bits as usual, it fills lpbi->bmiColors as well.
Question in EDIT section:
If lpvBits is NULL and lpvBits->bmiHeader.biBitCount is initialized to
zero, GetDIBits() fills in lpbi->bmiHeader only and lpbi->bmiColors is
not modified.
Is this the correct way to understand it?
Yes, that is correct.
And if so, is there a way to get GetDIBits() to fill in
lpbi->bmiColors, such as initializing lpvBits->bmiHeader.biBitCount to
bit-depth of the bitmap?
Yes, there is a way to get the function to return the color table, but as explained in answer to Q2, initializing lpvBits->bmiHeader.biBitCount to bit-depth of the bitmap alone is not enough. All members of lpvBits->bmiHeader must be appropriately initialized and lpvBits must be non-null. This is basically the same as calling the function to get the DIB bits.

DirectX Partial Screen Capture

I am trying to create a program that will capture a full screen directx application, look for a specific set of pixels on the screen and if it finds it then draw an image on the screen.
I have been able to set up the application to capture the screen the directx libraries using the code the answer for this question Capture screen using DirectX
In this example the code saves to the harddrive using the IWIC libraries. I would rather manipulate the pixels instead of saving it.
After I have captured the screen and have a LPBYTE of the entire screen pixels I am unsure how to crop it to the region I want and then being able to manipulate the pixel array. Is it just a multi dimensional byte array?
The way I think I should do it is
Capture screen to IWIC bitmap (done).
Convert IWIC bitmap to ID2D1 bitmap using ID2D1RenderTarget::CreateBitmapFromWicBitmap
Create new ID2D1::Bitmap to store partial image.
Copy region of the ID2D1 bitmap to a new bitmap using ID2D1::CopyFromBitmap.
Render back onto screen using ID2D1 .
Any help on any of this would be so much appreciated.
Here is a modified version of the original code that only captures a portion of the screen into a buffer, and also gives back the stride. Then it browses all the pixels, dumps their colors as a sample usage of the returned buffer.
In this sample, the buffer is allocated by the function, so you must free it once you've used it:
// sample usage
int main()
{
LONG left = 10;
LONG top = 10;
LONG width = 100;
LONG height = 100;
LPBYTE buffer;
UINT stride;
RECT rc = { left, top, left + width, top + height };
Direct3D9TakeScreenshot(D3DADAPTER_DEFAULT, &buffer, &stride, &rc);
// In 32bppPBGRA format, each pixel is represented by 4 bytes
// with one byte each for blue, green, red, and the alpha channel, in that order.
// But don't forget this is all modulo endianness ...
// So, on Intel architecture, if we read a pixel from memory
// as a DWORD, it's reversed (ARGB). The macros below handle that.
// browse every pixel by line
for (int h = 0; h < height; h++)
{
LPDWORD pixels = (LPDWORD)(buffer + h * stride);
for (int w = 0; w < width; w++)
{
DWORD pixel = pixels[w];
wprintf(L"#%02X#%02X#%02X#%02X\n", GetBGRAPixelAlpha(pixel), GetBGRAPixelRed(pixel), GetBGRAPixelGreen(pixel), GetBGRAPixelBlue(pixel));
}
}
// get pixel at 50, 50 in the buffer, as #ARGB
DWORD pixel = GetBGRAPixel(buffer, stride, 50, 50);
wprintf(L"#%02X#%02X#%02X#%02X\n", GetBGRAPixelAlpha(pixel), GetBGRAPixelRed(pixel), GetBGRAPixelGreen(pixel), GetBGRAPixelBlue(pixel));
SavePixelsToFile32bppPBGRA(width, height, stride, buffer, L"test.png", GUID_ContainerFormatPng);
LocalFree(buffer);
return 0;;
}
#define GetBGRAPixelBlue(p) (LOBYTE(p))
#define GetBGRAPixelGreen(p) (HIBYTE(p))
#define GetBGRAPixelRed(p) (LOBYTE(HIWORD(p)))
#define GetBGRAPixelAlpha(p) (HIBYTE(HIWORD(p)))
#define GetBGRAPixel(b,s,x,y) (((LPDWORD)(((LPBYTE)b) + y * s))[x])
int main()
HRESULT Direct3D9TakeScreenshot(UINT adapter, LPBYTE *pBuffer, UINT *pStride, const RECT *pInputRc = nullptr)
{
if (!pBuffer || !pStride) return E_INVALIDARG;
HRESULT hr = S_OK;
IDirect3D9 *d3d = nullptr;
IDirect3DDevice9 *device = nullptr;
IDirect3DSurface9 *surface = nullptr;
D3DPRESENT_PARAMETERS parameters = { 0 };
D3DDISPLAYMODE mode;
D3DLOCKED_RECT rc;
*pBuffer = NULL;
*pStride = 0;
// init D3D and get screen size
d3d = Direct3DCreate9(D3D_SDK_VERSION);
HRCHECK(d3d->GetAdapterDisplayMode(adapter, &mode));
LONG width = pInputRc ? (pInputRc->right - pInputRc->left) : mode.Width;
LONG height = pInputRc ? (pInputRc->bottom - pInputRc->top) : mode.Height;
parameters.Windowed = TRUE;
parameters.BackBufferCount = 1;
parameters.BackBufferHeight = height;
parameters.BackBufferWidth = width;
parameters.SwapEffect = D3DSWAPEFFECT_DISCARD;
parameters.hDeviceWindow = NULL;
// create device & capture surface (note it needs desktop size, not our capture size)
HRCHECK(d3d->CreateDevice(adapter, D3DDEVTYPE_HAL, NULL, D3DCREATE_SOFTWARE_VERTEXPROCESSING, &parameters, &device));
HRCHECK(device->CreateOffscreenPlainSurface(mode.Width, mode.Height, D3DFMT_A8R8G8B8, D3DPOOL_SYSTEMMEM, &surface, nullptr));
// get pitch/stride to compute the required buffer size
HRCHECK(surface->LockRect(&rc, pInputRc, 0));
*pStride = rc.Pitch;
HRCHECK(surface->UnlockRect());
// allocate buffer
*pBuffer = (LPBYTE)LocalAlloc(0, *pStride * height);
if (!*pBuffer)
{
hr = E_OUTOFMEMORY;
goto cleanup;
}
// get the data
HRCHECK(device->GetFrontBufferData(0, surface));
// copy it into our buffer
HRCHECK(surface->LockRect(&rc, pInputRc, 0));
CopyMemory(*pBuffer, rc.pBits, rc.Pitch * height);
HRCHECK(surface->UnlockRect());
cleanup:
if (FAILED(hr))
{
if (*pBuffer)
{
LocalFree(*pBuffer);
*pBuffer = NULL;
}
*pStride = 0;
}
RELEASE(surface);
RELEASE(device);
RELEASE(d3d);
return hr;
}

What are pixel size limits of GDI bitmaps/DCs in 32-bit and 64-bit processes?

I'm coding a Win32 application that performs low-level printing, for which I'm dealing with GDI bitmaps and device contexts.
The code basically does this (pseudo-code):
HDC hCompDc = ::CreateCompatibleDC(hDC);
BITMAPINFOHEADER infoHeader = {0};
infoHeader.biSize = sizeof(infoHeader);
infoHeader.biWidth = nWidth;
infoHeader.biHeight = -nHeight; //The document must be right side up
infoHeader.biPlanes = 1;
infoHeader.biBitCount = 24;
infoHeader.biCompression = BI_RGB;
BITMAPINFO info;
info.bmiHeader = infoHeader;
//Use file on disk to store large bitmap data
HANDLE hFileScatch = ::CreateFile(strTempPath,
GENERIC_READ | GENERIC_WRITE, 0, NULL,
CREATE_ALWAYS, FILE_ATTRIBUTE_NORMAL, NULL);
UINT64 ncbLineSz = nWidth * 3;
if(ncbLineSz & 0x3)
ncbLineSz = (ncbLineSz & ~0x3) + 0x4; //Must be DWORD aligned
UINT64 uiSzBmp = ncbLineSz * nHeight;
HANDLE hFileMapObj = ::CreateFileMapping(hFileScatch,
NULL, PAGE_READWRITE, (DWORD)(uiSzBmp >> 32), (DWORD)(uiSzBmp), NULL);
BYTE* pMemory = 0;
HBITMAP hBitmap = ::CreateDIBSection(hDC, &info, DIB_RGB_COLORS, (void**)&pMemory, hFileMapObj, 0);
::SelectObject(hCompDc, hBitmap);
//Do drawing on `hCompDc` using other GDI APIs
//...
//And print
::SetAbortProc(hPrintDC, _printerAbortProc);
::StartDoc(hPrintDC, &docInfo);
::StartPage(hPrintDC);
::BitBlt(hPrintDC,
rcPrintArea.left,
rcPrintArea.top,
rcRenderArea.Width(),
rcRenderArea.Height(),
hCompDc, //DC from
rcRenderArea.left,
rcRenderArea.top,
SRCCOPY);
::EndPage(hPrintDC);
//Clean up, etc.
//...
This works but to a certain degree. Let me explain:
Since I'm dealing with printer resolutions running at 300 dpi (vs. 96 dpi for a usual screen) the sizes of bitmaps turn out pretty large.
For instance, if I have a bitmap that is 3900x86625 pixels, it requires a DIB section of around 966 MB. (I can learn that if I substitute default memory allocation in CreateDIBSection in my code above with a file mapping object backed by a file on disk.) Otherwise, I understand that I can hit a limit of contiguous RAM space in a 32-bit process with that chunk of memory. But that is not what happens.
When I run my code with that size of a bitmap in a 32-bit process, the bitmap is created OK using file mapping object backed by a file on disk, but the resulting bitmap/image comes out completely blank, while the exact same code compiled in a 64-bit process, produces a normal looking bitmap.
So in light of that I was curious if there are specific image size restrictions that those GDI objects have? For 32-bit and 64-bit processes.
And also a side question -- is there any way to know that such size limit was reached while drawing?
PS. My actual code has all error checks in place for all Win32 APIs, and while the 32-bit process that I described above produced a blank bitmap, it did not fail in a single GDI API.

Editing bitmap in memory using C++

I have a two dimensional array of data that I want to display as an image.
The plan goes something like this -
Create a bitmap using CreateCompatibleBitmap (this results in a solid black bitmap and I can display this with no problems)
Edit the pixels of this bitmap to match my data
BitBlt the bitmap to the window
I think that I need a pointer to the place in memory where the pixel data begins. I've tried many different methods of doing this and googled it for 3 days and still haven't been able to even edit a single pixel.
Using a loop of SetPixel(HDC, x, y, Color) to set each pixel works but VERY slowly.
I have accomplished this in C# by locking the bitmap and editing the bits, but I am new to C++ and can't seem to figure out how to do something similar.
I have mostly been trying to use memset(p, value, length)
For "p" I have tried using the handle returned from CreateCompatibleBitmap, the DC for the bitmap, and the DC for the window. I have tried all sorts of values for the value and length.
I'm not sure if this is the right thing to use though.
I don't have to use a bitmap, that's just the only thing I know to do. Actually it would be awesome to find a way to directly change the main window's DC.
I do want to avoid libraries though. I am doing this purely for learning C++.
This took QUITE a bit of research so I'll post exactly how it is done for anyone else who may be looking.
This colors every pixel red.
hDC = BeginPaint(hWnd, &Ps);
const int
width = 400,
height = 400,
size = width * height * 3;
byte * data;
data = new byte[size];
for (int i = 0; i < size; i += 3)
{
data[i] = 0;
data[i + 1] = 0;
data[i + 2] = 255;
}
BITMAPINFOHEADER bmih;
bmih.biBitCount = 24;
bmih.biClrImportant = 0;
bmih.biClrUsed = 0;
bmih.biCompression = BI_RGB;
bmih.biWidth = width;
bmih.biHeight = height;
bmih.biPlanes = 1;
bmih.biSize = 40;
bmih.biSizeImage = size;
BITMAPINFO bmpi;
bmpi.bmiHeader = bmih;
SetDIBitsToDevice(hDC, 0, 0, width, height, 0, 0, 0, height, data, &bmpi, DIB_RGB_COLORS);
delete[] data;
memset can be used on the actually RGB information array (but you need to also know the format of the bitmap, if a pixel has 32 or 24 bits ).
From a bit of research on msdn, it seems that what you want to get is the BITMAP structure :
http://msdn.microsoft.com/en-us/library/k1sf4cx2.aspx
There you have the bmBits on which you can memset.
How to get there from your function ?
Well, CreateCompatibleBitmap returns a HBITMAP structure and it seems you can get BITMAP from HBITMAP with the following code :
BITMAP bmp;
GetObject(hBmp, sizeof(BITMAP), &bmp);
This however seems to get a you copy of the existing bitmap info, which only solves your memset problem (you can now set the bitmap information with memset, eventhou I don't see any other use for memeset besides making the bmp all white or black).
There should be a function that allows you to set the DC bites to a bitmap thou, so you should be able to use the new bitmap as a parameter.

Creating 8bpp bitmap with GDI and saving it as a file

I have a perfectly working code that creates 32bpp bitmap and I need to change it so that 8bpp bitmap is created.
Here's the piece of code that creates 32bpp bitmap, draws into it, then it creates a bitmap file and store it into the vector of bytes:
// prepare bitmap:
BYTE* bitmap_data = NULL;
HDC hDC = GetDC(NULL);
HDC memHDC = CreateCompatibleDC(hDC);
BITMAPINFO bmi;
memset(&bmi, 0, sizeof(BITMAPINFO));
bmi.bmiHeader.biSize = sizeof(BITMAPINFOHEADER);
bmi.bmiHeader.biWidth = desiredWidth; // desiredWidth is 800
bmi.bmiHeader.biHeight = desiredHeight; // desiredHeight is 202
bmi.bmiHeader.biPlanes = 1;
bmi.bmiHeader.biBitCount = 32;
bmi.bmiHeader.biCompression = BI_RGB;
bmi.bmiHeader.biSizeImage = (((desiredWidth * bmi.bmiHeader.biBitCount + 31) & ~31) >> 3) * desiredHeight;
HBITMAP bitmap = CreateDIBSection(hDC, &bmi, DIB_RGB_COLORS, (void**)&bitmap_data, NULL, NULL);
ReleaseDC(NULL, hDC);
DeleteDC(hDC);
... // drawing into bitmap
// prepare bitmap file header:
BITMAPFILEHEADER bf;
memset(&bf, 0, sizeof(BITMAPFILEHEADER));
bf.bfType = MAKEWORD('B', 'M');
bf.bfOffBits = sizeof(BITMAPFILEHEADER) + bmi.bmiHeader.biSize;
bf.bfSize = bf.bfOffBits + bmi.bmiHeader.biSizeImage;
// write bitmap file into the vector:
std::vector<BYTE> bitmapData;
bitmapData.insert(bitmapData.end(), (BYTE*)&bf, ((BYTE*)&bf) + sizeof(BITMAPFILEHEADER));
bitmapData.insert(bitmapData.end(), (BYTE*)&bmi.bmiHeader, ((BYTE*)&bmi.bmiHeader) + sizeof(BITMAPINFOHEADER));
bitmapData.insert(bitmapData.end(), bitmap_data, bitmap_data + bmi.bmiHeader.biSizeImage);
And later the vector is stored into the file:
std::ofstream of("picture.bmp", std::ofstream::out | std::ofstream::binary);
of.write((char*)&bitmapData[0], bitmapData.size());
of.close();
and here's the output image:
What I've tried:
First step was naturally replacing 32 with 8 in this line: bmi.bmiHeader.biBitCount = 32; which resulted into image filled with solid grey colour. Then based on this answer I made following changes:
BITMAPINFO bmi;
memset(&bmi, 0, sizeof(BITMAPINFO));
changed into:
struct BITMAPINFO256 {
BITMAPINFOHEADER bmiHeader;
RGBQUAD bmiColors[256];
} bmi;
memset(&bmi, 0, sizeof(BITMAPINFO256));
added this loop right before CreateDIBSection is called:
for (UINT i = 0; i < 256; i++) {
bmi.bmiColors[i].rgbRed = i;
bmi.bmiColors[i].rgbGreen = i;
bmi.bmiColors[i].rgbBlue = i;
}
and when the bmi.bmiHeader is being written into the vector, the RGBQUAD array is included: so sizeof(BITMAPINFO256) expresses the size of the header.
The new code (full code here) produces this output:
Why the new image looks that way? What's going on there? What am I missing?
Any help will be appreciated.
It looks like an alignment problem. Make sure you update bfOffBits in the BITMAPFILEHEADER so that it points to the first byte of the pixel data. (If you don't change it, then it probably points to the beginning of the palette.)
In other words, sizeof(RGBQUAD)*256 should be added here as well:
bf.bfOffBits = sizeof(BITMAPFILEHEADER) + bmi.bmiHeader.biSize;
Also makes sure the first scanline starts on a DWORD boundary. That is, its offset from the beginning of the file should be a multiple of four bytes. Likewise, each scanline should be padded out to a multiple of four bytes. (You may not see these problems if your widths are nice even numbers. It's good to have an odd-width image among your test cases.)
You need to specify the size of the palette that you attached. Right now it's zero, so the palette is showing up as the first bunch of pixels in your image.
bmi.bmiHeader.biClrUsed = 256;
You need to generate a palette for your image.
Each pixel in a 32bit image is stored as 8-bits for Alpha, Red, Green and Blue.
Where as in a 8bit image, the value in each pixel an 8bit index into the palette.
Your for(i=0..255) { bmi.bmiColors[i].rgbRed = i; ....} code is generated an grey-scale palette.
If the whole image is coming out as grey then it sounds like an alignment error, from memory the width of a palettized image must be a multiple of 4.
Try saving a SMALL 256 colour (aka 8-bit image) from Paint and compare in a hex editor.