Abstract class for an image with unspecified, possibly compressed, memory format. The idea is that both uncompressed images (TCastleImage) and images compressed for GPU (TGPUCompressedImage) are derived from this class.
Operate on this by Get/Realloc/FreeMem. It's always freed and nil'ed in destructor.
Methods
destructor Destroy; override;
function Size: Cardinal; virtual; abstract;
Size of image contents in bytes.
function IsEmpty: boolean;
Is an image empty.
True means that RawPixels = Nil, and Width * Height * Depth = 0 (so either Width = 0 or Height = 0 or Depth = 0).
False means that RawPixels <> nil and Width * Height * Depth <> 0 (so all Width > 0 and Height > 0 and Depth > 0, since they are Cardinal (unsigned) always).
function HasAlpha: boolean; virtual;
Does an image have an alpha channel.
You may also be interested in the AlphaChannel. AlphaChannel answers always atNone if HasAlpha = false, and always acSimpleYesNo or acFullRange if HasAlpha = true. But AlphaChannel may perform longer analysis of pixels (to differ between acSimpleYesNo and acFullRange), while this function always executes ultra-fast (as it's constant for each TCastleImage descendant).
Descendants implementors notes: in this class, TCastleImage, this returns False. Override to return True for images with alpha channel.
Check does an image have an alpha channel, and if yes analyze alpha channel: is it a single yes-no (only full or none values), or does it have alpha values in between?
This is quite useful for automatic detection how alpha textures should be displayed: for simple yes/no alpha, OpenGL alpha_test is a simple solution. For full range alpha, OpenGL blending should be used. Blending is a little problematic, since it requires special rendering order, since it doesn't cooperate nicely with Z-buffer. That's why we try to detect simple yes/no alpha textures, so that we're able to use simpler alpha test for them.
We return "simple yes/no alpha channel" is all the alpha values (for every pixel) are 0, or 255, or (when AlphaTolerance <> 0) are close to them by AlphaTolerance. So, to be precise, alpha value must be <= AlphaTolerance, or >= 255 - AlphaTolerance. If any alpha value is between [AlphaTolerance + 1, 255 - AlphaTolerance - 1] then we return "full range alpha channel". Note that for AlphaTolerance >= 128, all images are treated as "simple yes/no alpha". Usually, you want to keep AlphaTolerance small.
Descendants implementors notes: in this class, this simply always returns atNone. For descendants that have alpha channel, implement it, honouring AlphaTolerance as described.