BMP Study

BMP file format study

Header Format

Assuming BITMAPINFOHEADER header format

+--+-+----------------------------+
|0 |2|magic number                | BMP file header
+--+-+----------------------------+
|2 |4|BMP size                    |
+--+-+----------------------------+
|6 |2|reserved                    |
+--+-+----------------------------+
|8 |2|reserved                    |
+--+-+----------------------------+
|10|4|starting address of image   |
+--+-+----------------------------+
|   BITMAPINFOHEADER exclusive    | BIB header
+--+-+----------------------------+
|14|4|size of this header         |
+--+-+----------------------------+
|18|4|width                       |
+--+-+----------------------------+
|22|4|height                      |
+--+-+----------------------------+
|26|2|number of color planes      |
+--+-+----------------------------+
|28|2|number of bits per pixel    |
+--+-+----------------------------+
|30|4|compression methd           |
+--+-+----------------------------+
|34|4|image size - raw bitmap data|
+--+-+----------------------------+
|38|4|horizontal resolution       |
+--+-+----------------------------+
|42|4|vertical resolution         |
+--+-+----------------------------+
|46|4|number of colors 2^n        |
+--+-+----------------------------+
|50|4|number of important colors  |
+--+-+----------------------------+

Notes

Header fields use the WORD and DWORD data types from Windows. A word is 16 bits, whilst dword is 32 bits, making them representable as 4 and 8 hex digits respectively.

Within xxd, this means every column is a word long, whilst 2 columns represent a dword.

Integer values are all stored in little-endian format.

Rows are padded up to a multiple of 4 bytes. Thus the formula for bits per row is:

(bitsPerPixel * imageWidth + 31) / 32 * 32

Or

$$ \left\lfloor \frac{\text{bits per pixel} \cdot \text{image width} + 31}{32} \right\rfloor \cdot 32 $$

Using the C code (bitsPerPixel * imageWidth + 31) & ~31, we can round up to the nearest 32 bits.

An ~31 int is equal to ...100000. An and operation causes the last 5 bits to be removed, effectively rounding down to the nearest multiple of 32.

Bitmap file header is 14 bytes, long. Use xxd -l 14 demo.bmp. The offset or starting of address of the pixel array is the last 4 bytes (2 columns).