path: root/plugins/FreeImage/Source/FreeImageToolkit/ClassicRotate.cpp

// ==========================================================
// Bitmap rotation by means of 3 shears.
//
// Design and implementation by
// - Hervé Drolon (drolon@infonie.fr)
// - Thorsten Radde (support@IdealSoftware.com)
// - Mihail Naydenov (mnaydenov@users.sourceforge.net)
//
// This file is part of FreeImage 3
//
// COVERED CODE IS PROVIDED UNDER THIS LICENSE ON AN "AS IS" BASIS, WITHOUT WARRANTY
// OF ANY KIND, EITHER EXPRESSED OR IMPLIED, INCLUDING, WITHOUT LIMITATION, WARRANTIES
// THAT THE COVERED CODE IS FREE OF DEFECTS, MERCHANTABLE, FIT FOR A PARTICULAR PURPOSE
// OR NON-INFRINGING. THE ENTIRE RISK AS TO THE QUALITY AND PERFORMANCE OF THE COVERED
// CODE IS WITH YOU. SHOULD ANY COVERED CODE PROVE DEFECTIVE IN ANY RESPECT, YOU (NOT
// THE INITIAL DEVELOPER OR ANY OTHER CONTRIBUTOR) ASSUME THE COST OF ANY NECESSARY
// SERVICING, REPAIR OR CORRECTION. THIS DISCLAIMER OF WARRANTY CONSTITUTES AN ESSENTIAL
// PART OF THIS LICENSE. NO USE OF ANY COVERED CODE IS AUTHORIZED HEREUNDER EXCEPT UNDER
// THIS DISCLAIMER.
//
// Use at your own risk!
// ==========================================================

/* 
 ============================================================
 References : 
 [1] Paeth A., A Fast Algorithm for General Raster Rotation. 
 Graphics Gems, p. 179, Andrew Glassner editor, Academic Press, 1990. 
 [2] Yariv E., High quality image rotation (rotate by shear). 
 [Online] http://www.codeproject.com/bitmap/rotatebyshear.asp
 [3] Treskunov A., Fast and high quality true-color bitmap rotation function.
 [Online] http://anton.treskunov.net/Software/doc/fast_and_high_quality_true_color_bitmap_rotation_function.html
 ============================================================
*/

#include "FreeImage.h"
#include "Utilities.h"

#define RBLOCK		64	// image blocks of RBLOCK*RBLOCK pixels

// --------------------------------------------------------------------------

/**
Skews a row horizontally (with filtered weights). 
Limited to 45 degree skewing only. Filters two adjacent pixels.
Parameter T can be BYTE, WORD of float. 
@param src Pointer to source image to rotate
@param dst Pointer to destination image
@param row Row index
@param iOffset Skew offset
@param dWeight Relative weight of right pixel
@param bkcolor Background color
*/
template <class T> void 
HorizontalSkewT(FIBITMAP *src, FIBITMAP *dst, int row, int iOffset, double weight, const void *bkcolor = NULL) {
	int iXPos;

	const unsigned src_width  = FreeImage_GetWidth(src);
	const unsigned dst_width  = FreeImage_GetWidth(dst);

	T pxlSrc[4], pxlLeft[4], pxlOldLeft[4];	// 4 = 4*sizeof(T) max
	
	// background
	const T pxlBlack[4] = {0, 0, 0, 0 };
	const T *pxlBkg = static_cast<const T*>(bkcolor); // assume at least bytespp and 4*sizeof(T) max
	if(!pxlBkg) {
		// default background color is black
		pxlBkg = pxlBlack;
	}

	// calculate the number of bytes per pixel
	const unsigned bytespp = FreeImage_GetLine(src) / FreeImage_GetWidth(src);
	// calculate the number of samples per pixel
	const unsigned samples = bytespp / sizeof(T);

	BYTE *src_bits = FreeImage_GetScanLine(src, row);
	BYTE *dst_bits = FreeImage_GetScanLine(dst, row);

	// fill gap left of skew with background
	if(bkcolor) {
		for(int k = 0; k < iOffset; k++) {
			memcpy(&dst_bits[k * bytespp], bkcolor, bytespp);
		}
		AssignPixel((BYTE*)&pxlOldLeft[0], (BYTE*)bkcolor, bytespp);
	} else {
		if(iOffset > 0) {
			memset(dst_bits, 0, iOffset * bytespp);
		}		
		memset(&pxlOldLeft[0], 0, bytespp);
	}

	for(unsigned i = 0; i < src_width; i++) {
		// loop through row pixels
		AssignPixel((BYTE*)&pxlSrc[0], (BYTE*)src_bits, bytespp);
		// calculate weights
		for(unsigned j = 0; j < samples; j++) {
			pxlLeft[j] = static_cast<T>(pxlBkg[j] + (pxlSrc[j] - pxlBkg[j]) * weight + 0.5);
		}
		// check boundaries 
		iXPos = i + iOffset;
		if((iXPos >= 0) && (iXPos < (int)dst_width)) {
			// update left over on source
			for(unsigned j = 0; j < samples; j++) {
				pxlSrc[j] = pxlSrc[j] - (pxlLeft[j] - pxlOldLeft[j]);
			}
			AssignPixel((BYTE*)&dst_bits[iXPos*bytespp], (BYTE*)&pxlSrc[0], bytespp);
		}
		// save leftover for next pixel in scan
		AssignPixel((BYTE*)&pxlOldLeft[0], (BYTE*)&pxlLeft[0], bytespp);

		// next pixel in scan
		src_bits += bytespp;
	}			

	// go to rightmost point of skew
	iXPos = src_width + iOffset; 

	if((iXPos >= 0) && (iXPos < (int)dst_width)) {
		dst_bits = FreeImage_GetScanLine(dst, row) + iXPos * bytespp;

		// If still in image bounds, put leftovers there
		AssignPixel((BYTE*)dst_bits, (BYTE*)&pxlOldLeft[0], bytespp);

		// clear to the right of the skewed line with background
		dst_bits += bytespp;
		if(bkcolor) {
			for(unsigned i = 0; i < dst_width - iXPos - 1; i++) {
				memcpy(&dst_bits[i * bytespp], bkcolor, bytespp);
			}
		} else {
			memset(dst_bits, 0, bytespp * (dst_width - iXPos - 1));
		}

	}
}

/**
Skews a row horizontally (with filtered weights). 
Limited to 45 degree skewing only. Filters two adjacent pixels.
@param src Pointer to source image to rotate
@param dst Pointer to destination image
@param row Row index
@param iOffset Skew offset
@param dWeight Relative weight of right pixel
@param bkcolor Background color
*/
static void 
HorizontalSkew(FIBITMAP *src, FIBITMAP *dst, int row, int iOffset, double dWeight, const void *bkcolor) {
	FREE_IMAGE_TYPE image_type = FreeImage_GetImageType(src);

	switch(image_type) {
		case FIT_BITMAP:
			switch(FreeImage_GetBPP(src)) {
				case 8:
				case 24:
				case 32:
					HorizontalSkewT<BYTE>(src, dst, row, iOffset, dWeight, bkcolor);
				break;
			}
			break;
		case FIT_UINT16:
		case FIT_RGB16:
		case FIT_RGBA16:
			HorizontalSkewT<WORD>(src, dst, row, iOffset, dWeight, bkcolor);
			break;
		case FIT_FLOAT:
		case FIT_RGBF:
		case FIT_RGBAF:
			HorizontalSkewT<float>(src, dst, row, iOffset, dWeight, bkcolor);
			break;
	}
}

/**
Skews a column vertically (with filtered weights). 
Limited to 45 degree skewing only. Filters two adjacent pixels.
Parameter T can be BYTE, WORD of float. 
@param src Pointer to source image to rotate
@param dst Pointer to destination image
@param col Column index
@param iOffset Skew offset
@param dWeight Relative weight of upper pixel
@param bkcolor Background color
*/
template <class T> void 
VerticalSkewT(FIBITMAP *src, FIBITMAP *dst, int col, int iOffset, double weight, const void *bkcolor = NULL) {
	int iYPos;

	unsigned src_height = FreeImage_GetHeight(src);
	unsigned dst_height = FreeImage_GetHeight(dst);

	T pxlSrc[4], pxlLeft[4], pxlOldLeft[4];	// 4 = 4*sizeof(T) max

	// background
	const T pxlBlack[4] = {0, 0, 0, 0 };
	const T *pxlBkg = static_cast<const T*>(bkcolor); // assume at least bytespp and 4*sizeof(T) max
	if(!pxlBkg) {
		// default background color is black
		pxlBkg = pxlBlack;
	}

	// calculate the number of bytes per pixel
	const unsigned bytespp = FreeImage_GetLine(src) / FreeImage_GetWidth(src);
	// calculate the number of samples per pixel
	const unsigned samples = bytespp / sizeof(T);

	const unsigned src_pitch = FreeImage_GetPitch(src);
	const unsigned dst_pitch = FreeImage_GetPitch(dst);
	const unsigned index = col * bytespp;

	BYTE *src_bits = FreeImage_GetBits(src) + index;
	BYTE *dst_bits = FreeImage_GetBits(dst) + index;

	// fill gap above skew with background
	if(bkcolor) {
		for(int k = 0; k < iOffset; k++) {
			memcpy(dst_bits, bkcolor, bytespp);
			dst_bits += dst_pitch;
		}
		memcpy(&pxlOldLeft[0], bkcolor, bytespp);
	} else {
		for(int k = 0; k < iOffset; k++) {
			memset(dst_bits, 0, bytespp);
			dst_bits += dst_pitch;
		}
		memset(&pxlOldLeft[0], 0, bytespp);
	}

	for(unsigned i = 0; i < src_height; i++) {
		// loop through column pixels
		AssignPixel((BYTE*)(&pxlSrc[0]), src_bits, bytespp);
		// calculate weights
		for(unsigned j = 0; j < samples; j++) {
			pxlLeft[j] = static_cast<T>(pxlBkg[j] + (pxlSrc[j] - pxlBkg[j]) * weight + 0.5);
		}
		// check boundaries
		iYPos = i + iOffset;
		if((iYPos >= 0) && (iYPos < (int)dst_height)) {
			// update left over on source
			for(unsigned j = 0; j < samples; j++) {
				pxlSrc[j] = pxlSrc[j] - (pxlLeft[j] - pxlOldLeft[j]);
			}
			dst_bits = FreeImage_GetScanLine(dst, iYPos) + index;
			AssignPixel(dst_bits, (BYTE*)(&pxlSrc[0]), bytespp);
		}
		// save leftover for next pixel in scan
		AssignPixel((BYTE*)(&pxlOldLeft[0]), (BYTE*)(&pxlLeft[0]), bytespp);

		// next pixel in scan
		src_bits += src_pitch;
	}
	// go to bottom point of skew
	iYPos = src_height + iOffset;

	if((iYPos >= 0) && (iYPos < (int)dst_height)) {
		dst_bits = FreeImage_GetScanLine(dst, iYPos) + index;

		// if still in image bounds, put leftovers there				
		AssignPixel((BYTE*)(dst_bits), (BYTE*)(&pxlOldLeft[0]), bytespp);

		// clear below skewed line with background
		if(bkcolor) {
			while(++iYPos < (int)dst_height) {					
				dst_bits += dst_pitch;
				AssignPixel((BYTE*)(dst_bits), (BYTE*)(bkcolor), bytespp);
			}
		} else {
			while(++iYPos < (int)dst_height) {					
				dst_bits += dst_pitch;
				memset(dst_bits, 0, bytespp);
			}
		}
	}
}

/**
Skews a column vertically (with filtered weights). 
Limited to 45 degree skewing only. Filters two adjacent pixels.
@param src Pointer to source image to rotate
@param dst Pointer to destination image
@param col Column index
@param iOffset Skew offset
@param dWeight Relative weight of upper pixel
@param bkcolor Background color
*/
static void 
VerticalSkew(FIBITMAP *src, FIBITMAP *dst, int col, int iOffset, double dWeight, const void *bkcolor) {
	FREE_IMAGE_TYPE image_type = FreeImage_GetImageType(src);

	switch(image_type) {
		case FIT_BITMAP:
			switch(FreeImage_GetBPP(src)) {
				case 8:
				case 24:
				case 32:
					VerticalSkewT<BYTE>(src, dst, col, iOffset, dWeight, bkcolor);
					break;
			}
			break;
			case FIT_UINT16:
			case FIT_RGB16:
			case FIT_RGBA16:
				VerticalSkewT<WORD>(src, dst, col, iOffset, dWeight, bkcolor);
				break;
			case FIT_FLOAT:
			case FIT_RGBF:
			case FIT_RGBAF:
				VerticalSkewT<float>(src, dst, col, iOffset, dWeight, bkcolor);
				break;
	}
} 

/**
Rotates an image by 90 degrees (counter clockwise). 
Precise rotation, no filters required.<br>
Code adapted from CxImage (http://www.xdp.it/cximage.htm)
@param src Pointer to source image to rotate
@return Returns a pointer to a newly allocated rotated image if successful, returns NULL otherwise
*/
static FIBITMAP* 
Rotate90(FIBITMAP *src) {

	const unsigned bpp = FreeImage_GetBPP(src);

	const unsigned src_width  = FreeImage_GetWidth(src);
	const unsigned src_height = FreeImage_GetHeight(src);	
	const unsigned dst_width  = src_height;
	const unsigned dst_height = src_width;

	FREE_IMAGE_TYPE image_type = FreeImage_GetImageType(src);

	// allocate and clear dst image
	FIBITMAP *dst = FreeImage_AllocateT(image_type, dst_width, dst_height, bpp);
	if(NULL == dst) return NULL;

	// get src and dst scan width
	const unsigned src_pitch  = FreeImage_GetPitch(src);
	const unsigned dst_pitch  = FreeImage_GetPitch(dst);

	switch(image_type) {
		case FIT_BITMAP:
			if(bpp == 1) {
				// speedy rotate for BW images

				BYTE *bsrc  = FreeImage_GetBits(src); 
				BYTE *bdest = FreeImage_GetBits(dst);

				BYTE *dbitsmax = bdest + dst_height * dst_pitch - 1;

				for(unsigned y = 0; y < src_height; y++) {
					// figure out the column we are going to be copying to
					const div_t div_r = div(y, 8);
					// set bit pos of src column byte
					const BYTE bitpos = (BYTE)(128 >> div_r.rem);
					BYTE *srcdisp = bsrc + y * src_pitch;
					for(unsigned x = 0; x < src_pitch; x++) {
						// get source bits
						BYTE *sbits = srcdisp + x;
						// get destination column
						BYTE *nrow = bdest + (dst_height - 1 - (x * 8)) * dst_pitch + div_r.quot;
						for (int z = 0; z < 8; z++) {
						   // get destination byte
							BYTE *dbits = nrow - z * dst_pitch;
							if ((dbits < bdest) || (dbits > dbitsmax)) break;
							if (*sbits & (128 >> z)) *dbits |= bitpos;
						}
					}
				}
			}
			else if((bpp == 8) || (bpp == 24) || (bpp == 32)) {
				// anything other than BW :
				// This optimized version of rotation rotates image by smaller blocks. It is quite
				// a bit faster than obvious algorithm, because it produces much less CPU cache misses.
				// This optimization can be tuned by changing block size (RBLOCK). 96 is good value for current
				// CPUs (tested on Athlon XP and Celeron D). Larger value (if CPU has enough cache) will increase
				// speed somehow, but once you drop out of CPU's cache, things will slow down drastically.
				// For older CPUs with less cache, lower value would yield better results.

				BYTE *bsrc  = FreeImage_GetBits(src);  // source pixels
				BYTE *bdest = FreeImage_GetBits(dst);  // destination pixels

				// calculate the number of bytes per pixel (1 for 8-bit, 3 for 24-bit or 4 for 32-bit)
				const unsigned bytespp = FreeImage_GetLine(src) / FreeImage_GetWidth(src);
				
				// for all image blocks of RBLOCK*RBLOCK pixels
				
				// x-segment
				for(unsigned xs = 0; xs < dst_width; xs += RBLOCK) {
					// y-segment
					for(unsigned ys = 0; ys < dst_height; ys += RBLOCK) {
						for(unsigned y = ys; y < MIN(dst_height, ys + RBLOCK); y++) {    // do rotation
							const unsigned y2 = dst_height - y - 1;
							// point to src pixel at (y2, xs)
							BYTE *src_bits = bsrc + (xs * src_pitch) + (y2 * bytespp);
							// point to dst pixel at (xs, y)
							BYTE *dst_bits = bdest + (y * dst_pitch) + (xs * bytespp);
							for(unsigned x = xs; x < MIN(dst_width, xs + RBLOCK); x++) {
								// dst.SetPixel(x, y, src.GetPixel(y2, x));
								AssignPixel(dst_bits, src_bits, bytespp);
								dst_bits += bytespp;
								src_bits += src_pitch;
							}
						}
					}
				}
			}
			break;
		case FIT_UINT16:
		case FIT_RGB16:
		case FIT_RGBA16:
		case FIT_FLOAT:
		case FIT_RGBF:
		case FIT_RGBAF:
		{
			BYTE *bsrc  = FreeImage_GetBits(src);  // source pixels
			BYTE *bdest = FreeImage_GetBits(dst);  // destination pixels

			// calculate the number of bytes per pixel
			const unsigned bytespp = FreeImage_GetLine(src) / FreeImage_GetWidth(src);

			for(unsigned y = 0; y < dst_height; y++) {
				BYTE *src_bits = bsrc + (src_width - 1 - y) * bytespp;
				BYTE *dst_bits = bdest + (y * dst_pitch);
				for(unsigned x = 0; x < dst_width; x++) {
					AssignPixel(dst_bits, src_bits, bytespp);
					src_bits += src_pitch;
					dst_bits += bytespp;
				}
			}
		}
		break;
	}

	return dst;
}

/**
Rotates an image by 180 degrees (counter clockwise). 
Precise rotation, no filters required.
@param src Pointer to source image to rotate
@return Returns a pointer to a newly allocated rotated image if successful, returns NULL otherwise
*/
static FIBITMAP* 
Rotate180(FIBITMAP *src) {
	int x, y, k, pos;

	const int bpp = FreeImage_GetBPP(src);

	const int src_width  = FreeImage_GetWidth(src);
	const int src_height = FreeImage_GetHeight(src);
	const int dst_width  = src_width;
	const int dst_height = src_height;

	FREE_IMAGE_TYPE image_type = FreeImage_GetImageType(src);

	FIBITMAP *dst = FreeImage_AllocateT(image_type, dst_width, dst_height, bpp);
	if(NULL == dst) return NULL;

	switch(image_type) {
		case FIT_BITMAP:
			if(bpp == 1) {
				for(int y = 0; y < src_height; y++) {
					BYTE *src_bits = FreeImage_GetScanLine(src, y);
					BYTE *dst_bits = FreeImage_GetScanLine(dst, dst_height - y - 1);
					for(int x = 0; x < src_width; x++) {
						// get bit at (x, y)
						k = (src_bits[x >> 3] & (0x80 >> (x & 0x07))) != 0;
						// set bit at (dst_width - x - 1, dst_height - y - 1)
						pos = dst_width - x - 1;
						k ? dst_bits[pos >> 3] |= (0x80 >> (pos & 0x7)) : dst_bits[pos >> 3] &= (0xFF7F >> (pos & 0x7));
					}			
				}
				break;
			}
			// else if((bpp == 8) || (bpp == 24) || (bpp == 32)) FALL TROUGH
		case FIT_UINT16:
		case FIT_RGB16:
		case FIT_RGBA16:
		case FIT_FLOAT:
		case FIT_RGBF:
		case FIT_RGBAF:
		{
			 // Calculate the number of bytes per pixel
			const int bytespp = FreeImage_GetLine(src) / FreeImage_GetWidth(src);

			for(y = 0; y < src_height; y++) {
				BYTE *src_bits = FreeImage_GetScanLine(src, y);
				BYTE *dst_bits = FreeImage_GetScanLine(dst, dst_height - y - 1) + (dst_width - 1) * bytespp;
				for(x = 0; x < src_width; x++) {
					// get pixel at (x, y)
					// set pixel at (dst_width - x - 1, dst_height - y - 1)
					AssignPixel(dst_bits, src_bits, bytespp);
					src_bits += bytespp;
					dst_bits -= bytespp;					
				}				
			}
		}
		break;
	}

	return dst;
}

/**
Rotates an image by 270 degrees (counter clockwise). 
Precise rotation, no filters required.<br>
Code adapted from CxImage (http://www.xdp.it/cximage.htm)
@param src Pointer to source image to rotate
@return Returns a pointer to a newly allocated rotated image if successful, returns NULL otherwise
*/
static FIBITMAP* 
Rotate270(FIBITMAP *src) {
	int x2, dlineup;

	const unsigned bpp = FreeImage_GetBPP(src);

	const unsigned src_width  = FreeImage_GetWidth(src);
	const unsigned src_height = FreeImage_GetHeight(src);	
	const unsigned dst_width  = src_height;
	const unsigned dst_height = src_width;

	FREE_IMAGE_TYPE image_type = FreeImage_GetImageType(src);

	// allocate and clear dst image
	FIBITMAP *dst = FreeImage_AllocateT(image_type, dst_width, dst_height, bpp);
	if(NULL == dst) return NULL;

	// get src and dst scan width
	const unsigned src_pitch  = FreeImage_GetPitch(src);
	const unsigned dst_pitch  = FreeImage_GetPitch(dst);
	
	switch(image_type) {
		case FIT_BITMAP:
			if(bpp == 1) {
				// speedy rotate for BW images
				
				BYTE *bsrc  = FreeImage_GetBits(src); 
				BYTE *bdest = FreeImage_GetBits(dst);
				BYTE *dbitsmax = bdest + dst_height * dst_pitch - 1;
				dlineup = 8 * dst_pitch - dst_width;

				for(unsigned y = 0; y < src_height; y++) {
					// figure out the column we are going to be copying to
					const div_t div_r = div(y + dlineup, 8);
					// set bit pos of src column byte
					const BYTE bitpos = (BYTE)(1 << div_r.rem);
					const BYTE *srcdisp = bsrc + y * src_pitch;
					for(unsigned x = 0; x < src_pitch; x++) {
						// get source bits
						const BYTE *sbits = srcdisp + x;
						// get destination column
						BYTE *nrow = bdest + (x * 8) * dst_pitch + dst_pitch - 1 - div_r.quot;
						for(unsigned z = 0; z < 8; z++) {
						   // get destination byte
							BYTE *dbits = nrow + z * dst_pitch;
							if ((dbits < bdest) || (dbits > dbitsmax)) break;
							if (*sbits & (128 >> z)) *dbits |= bitpos;
						}
					}
				}
			} 
			else if((bpp == 8) || (bpp == 24) || (bpp == 32)) {
				// anything other than BW :
				// This optimized version of rotation rotates image by smaller blocks. It is quite
				// a bit faster than obvious algorithm, because it produces much less CPU cache misses.
				// This optimization can be tuned by changing block size (RBLOCK). 96 is good value for current
				// CPUs (tested on Athlon XP and Celeron D). Larger value (if CPU has enough cache) will increase
				// speed somehow, but once you drop out of CPU's cache, things will slow down drastically.
				// For older CPUs with less cache, lower value would yield better results.

				BYTE *bsrc  = FreeImage_GetBits(src);  // source pixels
				BYTE *bdest = FreeImage_GetBits(dst);  // destination pixels

				// Calculate the number of bytes per pixel (1 for 8-bit, 3 for 24-bit or 4 for 32-bit)
				const unsigned bytespp = FreeImage_GetLine(src) / FreeImage_GetWidth(src);

				// for all image blocks of RBLOCK*RBLOCK pixels

				// x-segment
				for(unsigned xs = 0; xs < dst_width; xs += RBLOCK) {
					// y-segment
					for(unsigned ys = 0; ys < dst_height; ys += RBLOCK) {
						for(unsigned x = xs; x < MIN(dst_width, xs + RBLOCK); x++) {    // do rotation
							x2 = dst_width - x - 1;
							// point to src pixel at (ys, x2)
							BYTE *src_bits = bsrc + (x2 * src_pitch) + (ys * bytespp);
							// point to dst pixel at (x, ys)
							BYTE *dst_bits = bdest + (ys * dst_pitch) + (x * bytespp);
							for(unsigned y = ys; y < MIN(dst_height, ys + RBLOCK); y++) {
								// dst.SetPixel(x, y, src.GetPixel(y, x2));
								AssignPixel(dst_bits, src_bits, bytespp);
								src_bits += bytespp;
								dst_bits += dst_pitch;
							}
						}
					}
				}
			}
			break;
		case FIT_UINT16:
		case FIT_RGB16:
		case FIT_RGBA16:
		case FIT_FLOAT:
		case FIT_RGBF:
		case FIT_RGBAF:
		{
			BYTE *bsrc  = FreeImage_GetBits(src);  // source pixels
			BYTE *bdest = FreeImage_GetBits(dst);  // destination pixels

			// calculate the number of bytes per pixel
			const unsigned bytespp = FreeImage_GetLine(src) / FreeImage_GetWidth(src);

			for(unsigned y = 0; y < dst_height; y++) {
				BYTE *src_bits = bsrc + (src_height - 1) * src_pitch + y * bytespp;
				BYTE *dst_bits = bdest + (y * dst_pitch);
				for(unsigned x = 0; x < dst_width; x++) {
					AssignPixel(dst_bits, src_bits, bytespp);
					src_bits -= src_pitch;
					dst_bits += bytespp;
				}
			}
		}
		break;
	}

	return dst;
}

/**
Rotates an image by a given degree in range [-45 .. +45] (counter clockwise) 
using the 3-shear technique.
@param src Pointer to source image to rotate
@param dAngle Rotation angle
@return Returns a pointer to a newly allocated rotated image if successful, returns NULL otherwise
*/
static FIBITMAP* 
Rotate45(FIBITMAP *src, double dAngle, const void *bkcolor) {
	const double ROTATE_PI = double(3.1415926535897932384626433832795);

	unsigned u;

	const unsigned bpp = FreeImage_GetBPP(src);

	const double dRadAngle = dAngle * ROTATE_PI / double(180); // Angle in radians
	const double dSinE = sin(dRadAngle);
	const double dTan = tan(dRadAngle / 2);

	const unsigned src_width  = FreeImage_GetWidth(src);
	const unsigned src_height = FreeImage_GetHeight(src);

	FREE_IMAGE_TYPE image_type = FreeImage_GetImageType(src);

	// Calc first shear (horizontal) destination image dimensions 
	const unsigned width_1  = src_width + unsigned((double)src_height * fabs(dTan) + 0.5);
	const unsigned height_1 = src_height; 

	// Perform 1st shear (horizontal)
	// ----------------------------------------------------------------------

	// Allocate image for 1st shear
	FIBITMAP *dst1 = FreeImage_AllocateT(image_type, width_1, height_1, bpp);
	if(NULL == dst1) {
		return NULL;
	}
	
	for(u = 0; u < height_1; u++) {  
		double dShear;

		if(dTan >= 0)	{
			// Positive angle
			dShear = (u + 0.5) * dTan;
		}
		else {
			// Negative angle
			dShear = (double(u) - height_1 + 0.5) * dTan;
		}
		int iShear = int(floor(dShear));
		HorizontalSkew(src, dst1, u, iShear, dShear - double(iShear), bkcolor);
	}

	// Perform 2nd shear  (vertical)
	// ----------------------------------------------------------------------

	// Calc 2nd shear (vertical) destination image dimensions
	const unsigned width_2  = width_1;
	unsigned height_2 = unsigned((double)src_width * fabs(dSinE) + (double)src_height * cos(dRadAngle) + 0.5) + 1;

	// Allocate image for 2nd shear
	FIBITMAP *dst2 = FreeImage_AllocateT(image_type, width_2, height_2, bpp);
	if(NULL == dst2) {
		FreeImage_Unload(dst1);
		return NULL;
	}

	double dOffset;     // Variable skew offset
	if(dSinE > 0)	{   
		// Positive angle
		dOffset = (src_width - 1.0) * dSinE;
	}
	else {
		// Negative angle
		dOffset = -dSinE * (double(src_width) - width_2);
	}

	for(u = 0; u < width_2; u++, dOffset -= dSinE) {
		int iShear = int(floor(dOffset));
		VerticalSkew(dst1, dst2, u, iShear, dOffset - double(iShear), bkcolor);
	}

	// Perform 3rd shear (horizontal)
	// ----------------------------------------------------------------------

	// Free result of 1st shear
	FreeImage_Unload(dst1);

	// Calc 3rd shear (horizontal) destination image dimensions
	const unsigned width_3  = unsigned(double(src_height) * fabs(dSinE) + double(src_width) * cos(dRadAngle) + 0.5) + 1;
	const unsigned height_3 = height_2;

	// Allocate image for 3rd shear
	FIBITMAP *dst3 = FreeImage_AllocateT(image_type, width_3, height_3, bpp);
	if(NULL == dst3) {
		FreeImage_Unload(dst2);
		return NULL;
	}

	if(dSinE >= 0) {
		// Positive angle
		dOffset = (src_width - 1.0) * dSinE * -dTan;
	}
	else {
		// Negative angle
		dOffset = dTan * ( (src_width - 1.0) * -dSinE + (1.0 - height_3) );
	}
	for(u = 0; u < height_3; u++, dOffset += dTan) {
		int iShear = int(floor(dOffset));
		HorizontalSkew(dst2, dst3, u, iShear, dOffset - double(iShear), bkcolor);
	}
	// Free result of 2nd shear    
	FreeImage_Unload(dst2);

	// Return result of 3rd shear
	return dst3;      
}

/**
Rotates a 1-, 8-, 24- or 32-bit image by a given angle (given in degree). 
Angle is unlimited, except for 1-bit images (limited to integer multiples of 90 degree). 
3-shears technique is used.
@param src Pointer to source image to rotate
@param dAngle Rotation angle
@return Returns a pointer to a newly allocated rotated image if successful, returns NULL otherwise
*/
static FIBITMAP* 
RotateAny(FIBITMAP *src, double dAngle, const void *bkcolor) {
	if(NULL == src) {
		return NULL;
	}

	FIBITMAP *image = src;

	while(dAngle >= 360) {
		// Bring angle to range of (-INF .. 360)
		dAngle -= 360;
	}
	while(dAngle < 0) {
		// Bring angle to range of [0 .. 360) 
		dAngle += 360;
	}
	if((dAngle > 45) && (dAngle <= 135)) {
		// Angle in (45 .. 135] 
		// Rotate image by 90 degrees into temporary image,
		// so it requires only an extra rotation angle 
		// of -45 .. +45 to complete rotation.
		image = Rotate90(src);
		dAngle -= 90;
	}
	else if((dAngle > 135) && (dAngle <= 225)) { 
		// Angle in (135 .. 225] 
		// Rotate image by 180 degrees into temporary image,
		// so it requires only an extra rotation angle 
		// of -45 .. +45 to complete rotation.
		image = Rotate180(src);
		dAngle -= 180;
	}
	else if((dAngle > 225) && (dAngle <= 315)) { 
		// Angle in (225 .. 315] 
		// Rotate image by 270 degrees into temporary image,
		// so it requires only an extra rotation angle 
		// of -45 .. +45 to complete rotation.
		image = Rotate270(src);
		dAngle -= 270;
	}

	// If we got here, angle is in (-45 .. +45]

	if(NULL == image)	{
		// Failed to allocate middle image
		return NULL;
	}

	if(0 == dAngle) {
		if(image == src) {
			// Nothing to do ...
			return FreeImage_Clone(src);
		} else {
			// No more rotation needed
			return image;
		}
	}
	else {
		// Perform last rotation
		FIBITMAP *dst = Rotate45(image, dAngle, bkcolor);

		if(src != image) {
			// Middle image was required, free it now.
			FreeImage_Unload(image);
		}

		return dst;
	}
}

// ==========================================================

FIBITMAP *DLL_CALLCONV 
FreeImage_Rotate(FIBITMAP *dib, double angle, const void *bkcolor) {
	if(!FreeImage_HasPixels(dib)) return NULL;

	if(0 == angle) {
		return FreeImage_Clone(dib);
	}
	// DIB are stored upside down ...
	angle *= -1;

	try {
		unsigned bpp = FreeImage_GetBPP(dib);
		FREE_IMAGE_TYPE image_type = FreeImage_GetImageType(dib);
		
		switch(image_type) {
			case FIT_BITMAP:
				if(bpp == 1) {
					// only rotate for integer multiples of 90 degree
					if(fmod(angle, 90) != 0)
						return NULL;

					// perform the rotation
					FIBITMAP *dst = RotateAny(dib, angle, bkcolor);
					if(!dst) throw(1);

					// build a greyscale palette
					RGBQUAD *dst_pal = FreeImage_GetPalette(dst);
					if(FreeImage_GetColorType(dib) == FIC_MINISBLACK) {
						dst_pal[0].rgbRed = dst_pal[0].rgbGreen = dst_pal[0].rgbBlue = 0;
						dst_pal[1].rgbRed = dst_pal[1].rgbGreen = dst_pal[1].rgbBlue = 255;			
					} else {
						dst_pal[0].rgbRed = dst_pal[0].rgbGreen = dst_pal[0].rgbBlue = 255;
						dst_pal[1].rgbRed = dst_pal[1].rgbGreen = dst_pal[1].rgbBlue = 0;			
					}

					// copy metadata from src to dst
					FreeImage_CloneMetadata(dst, dib);

					return dst;
				}
				else if((bpp == 8) || (bpp == 24) || (bpp == 32)) {
					FIBITMAP *dst = RotateAny(dib, angle, bkcolor);
					if(!dst) throw(1);
					
					if(bpp == 8) {
						// copy original palette to rotated bitmap
						RGBQUAD *src_pal = FreeImage_GetPalette(dib);
						RGBQUAD *dst_pal = FreeImage_GetPalette(dst);
						memcpy(&dst_pal[0], &src_pal[0], 256 * sizeof(RGBQUAD));

						// copy transparency table 
						FreeImage_SetTransparencyTable(dst, FreeImage_GetTransparencyTable(dib), FreeImage_GetTransparencyCount(dib));

						// copy background color 
						RGBQUAD bkcolor; 
						if( FreeImage_GetBackgroundColor(dib, &bkcolor) ) {
							FreeImage_SetBackgroundColor(dst, &bkcolor); 
						}

					}

					// copy metadata from src to dst
					FreeImage_CloneMetadata(dst, dib);

					return dst;
				}
				break;
			case FIT_UINT16:
			case FIT_RGB16:
			case FIT_RGBA16:
			case FIT_FLOAT:
			case FIT_RGBF:
			case FIT_RGBAF:
			{
				FIBITMAP *dst = RotateAny(dib, angle, bkcolor);
				if(!dst) throw(1);

				// copy metadata from src to dst
				FreeImage_CloneMetadata(dst, dib);

				return dst;
			}
			break;
		}

	} catch(int) {
		return NULL;
	}

	return NULL;
}