Amebis/wxWidgets
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

[ 1537065 ] wxImage: Higher quality scaling/sampling

Browse Source 
git-svn-id: https://svn.wxwidgets.org/svn/wx/wxWidgets/trunk@41412 c3d73ce0-8a6f-49c7-b76d-6d57e0e08775
This commit is contained in:
Robert Roebling
2006-09-24 12:47:16 +00:00
parent 98de4ac1c5
commit 07aaa1a4b8
3 changed files with 533 additions and 55 deletions
Download Patch File Download Diff File Expand all files Collapse all files

509
src/common/image.cpp
View File

@@ -414,7 +414,7 @@ wxImage wxImage::ShrinkBy( int xFactor , int yFactor ) const
return image;
}
wxImage wxImage::Scale( int width, int height ) const
wxImage wxImage::Scale( int width, int height, int quality ) const
{
wxImage image;
@@ -429,64 +429,86 @@ wxImage wxImage::Scale( int width, int height ) const
wxCHECK_MSG( (old_height > 0) && (old_width > 0), image,
wxT("invalid old image size") );
if ( old_width % width == 0 && old_width >= width &&
old_height % height == 0 && old_height >= height )
// If the image's new width and height are the same as the original, no need to waste time or CPU cycles
if(old_width == width && old_height == height)
return *this;
// Scale the image (...or more appropriately, resample the image) using either the high-quality or normal method as specified
if(quality == wxIMAGE_QUALITY_HIGH)
{
return ShrinkBy( old_width / width , old_height / height ) ;
}
image.Create( width, height, false );
unsigned char *data = image.GetData();
wxCHECK_MSG( data, image, wxT("unable to create image") );
unsigned char *source_data = M_IMGDATA->m_data;
unsigned char *target_data = data;
unsigned char *source_alpha = 0 ;
unsigned char *target_alpha = 0 ;
if (M_IMGDATA->m_hasMask)
{
image.SetMaskColour( M_IMGDATA->m_maskRed,
M_IMGDATA->m_maskGreen,
M_IMGDATA->m_maskBlue );
}
else
{
source_alpha = M_IMGDATA->m_alpha ;
if ( source_alpha )
// We need to check whether we are downsampling or upsampling the image
if(width < old_width && height < old_height)
{
image.SetAlpha() ;
target_alpha = image.GetAlpha() ;
// Downsample the image using the box averaging method for best results
image = ResampleBox(width, height);
}
else
{
// For upsampling or other random/wierd image dimensions we'll use a bicubic b-spline scaling method
image = ResampleBicubic(width, height);
}
}
long x_delta = (old_width<<16) / width;
long y_delta = (old_height<<16) / height;
unsigned char* dest_pixel = target_data;
long y = 0;
for ( long j = 0; j < height; j++ )
else // Default scaling method == simple pixel replication
{
if ( old_width % width == 0 && old_width >= width &&
old_height % height == 0 && old_height >= height )
{
unsigned char* src_line = &source_data[(y>>16)*old_width*3];
unsigned char* src_alpha_line = source_alpha ? &source_alpha[(y>>16)*old_width] : 0 ;
return ShrinkBy( old_width / width , old_height / height ) ;
}
image.Create( width, height, false );
long x = 0;
for ( long i = 0; i < width; i++ )
unsigned char *data = image.GetData();
wxCHECK_MSG( data, image, wxT("unable to create image") );
unsigned char *source_data = M_IMGDATA->m_data;
unsigned char *target_data = data;
unsigned char *source_alpha = 0 ;
unsigned char *target_alpha = 0 ;
if (M_IMGDATA->m_hasMask)
{
unsigned char* src_pixel = &src_line[(x>>16)*3];
unsigned char* src_alpha_pixel = source_alpha ? &src_alpha_line[(x>>16)] : 0 ;
dest_pixel[0] = src_pixel[0];
dest_pixel[1] = src_pixel[1];
dest_pixel[2] = src_pixel[2];
dest_pixel += 3;
if ( source_alpha )
*(target_alpha++) = *src_alpha_pixel ;
x += x_delta;
image.SetMaskColour( M_IMGDATA->m_maskRed,
M_IMGDATA->m_maskGreen,
M_IMGDATA->m_maskBlue );
}
else
{
source_alpha = M_IMGDATA->m_alpha ;
if ( source_alpha )
{
image.SetAlpha() ;
target_alpha = image.GetAlpha() ;
}
}
y += y_delta;
long x_delta = (old_width<<16) / width;
long y_delta = (old_height<<16) / height;
unsigned char* dest_pixel = target_data;
long y = 0;
for ( long j = 0; j < height; j++ )
{
unsigned char* src_line = &source_data[(y>>16)*old_width*3];
unsigned char* src_alpha_line = source_alpha ? &source_alpha[(y>>16)*old_width] : 0 ;
long x = 0;
for ( long i = 0; i < width; i++ )
{
unsigned char* src_pixel = &src_line[(x>>16)*3];
unsigned char* src_alpha_pixel = source_alpha ? &src_alpha_line[(x>>16)] : 0 ;
dest_pixel[0] = src_pixel[0];
dest_pixel[1] = src_pixel[1];
dest_pixel[2] = src_pixel[2];
dest_pixel += 3;
if ( source_alpha )
*(target_alpha++) = *src_alpha_pixel ;
x += x_delta;
}
y += y_delta;
}
}
// In case this is a cursor, make sure the hotspot is scaled accordingly:
@@ -500,6 +522,393 @@ wxImage wxImage::Scale( int width, int height ) const
return image;
}
wxImage wxImage::ResampleBox(int width, int height) const
{
// This function implements a simple pre-blur/box averaging method for downsampling that gives reasonably smooth results
// To scale the image down we will need to gather a grid of pixels of the size of the scale factor in each direction
// and then do an averaging of the pixels.
wxImage ret_image(width, height, false);
double scale_factor_x = double(M_IMGDATA->m_width) / width;
double scale_factor_y = double(M_IMGDATA->m_height) / height;
// If we want good-looking results we need to pre-blur the image a bit first
wxImage src_image(*this);
src_image = src_image.BlurHorizontal(scale_factor_x / 2);
src_image = src_image.BlurVertical(scale_factor_y / 2);
unsigned char* src_data = src_image.GetData();
unsigned char* src_alpha = src_image.GetAlpha();
unsigned char* dst_data = ret_image.GetData();
unsigned char* dst_alpha = NULL;
if(src_alpha)
{
ret_image.SetAlpha();
dst_alpha = ret_image.GetAlpha();
}
int x, y, i, j;
int averaged_pixels, src_pixel_index, src_x, src_y;
double sum_r, sum_g, sum_b, sum_a;
for(y = 0; y < height; y++) // Destination image - Y direction
{
// Source pixel in the Y direction
src_y = y * scale_factor_y;
for(x = 0; x < width; x++) // Destination image - X direction
{
// Source pixel in the X direction
src_x = x * scale_factor_x;
// Box of pixels to average
averaged_pixels = 0;
sum_r = sum_g = sum_b = sum_a = 0.0;
for(j = src_y - scale_factor_y / 2 + 1; j <= int(src_y + scale_factor_y / 2); j++) // Y direction
{
// We don't care to average pixels that don't exist (edges)
if(j < 0 || j > M_IMGDATA->m_height)
continue;
for(i = src_x - scale_factor_x / 2 + 1; i <= int(src_x + scale_factor_x / 2); i++) // X direction
{
// Don't average edge pixels
if(i < 0 || i > M_IMGDATA->m_width)
continue;
// Calculate the actual index in our source pixels
src_pixel_index = src_y * M_IMGDATA->m_width + src_x;
sum_r += src_data[src_pixel_index * 3 + 0];
sum_g += src_data[src_pixel_index * 3 + 1];
sum_b += src_data[src_pixel_index * 3 + 2];
if(src_alpha)
sum_a += src_alpha[src_pixel_index];
averaged_pixels++;
}
}
// Calculate the average from the sum and number of averaged pixels
dst_data[0] = int(sum_r / averaged_pixels);
dst_data[1] = int(sum_g / averaged_pixels);
dst_data[2] = int(sum_b / averaged_pixels);
dst_data += 3;
if(src_alpha)
*dst_alpha++ = sum_a / averaged_pixels;
}
}
return ret_image;
}
// The following two local functions are for the B-spline weighting of the bicubic sampling algorithm
static inline double spline_cube(double value)
{
return value <= 0.0 ? 0.0 : value * value * value;
}
static inline double spline_weight(double value)
{
return (spline_cube(value + 2) - 4 * spline_cube(value + 1) + 6 * spline_cube(value) - 4 * spline_cube(value - 1)) / 6;
}
// This is the bicubic resampling algorithm
wxImage wxImage::ResampleBicubic(int width, int height) const
{
// This function implements a Bicubic B-Spline algorithm for resampling. This method is certainly a little slower than wxImage's default
// pixel replication method, however for most reasonably sized images not being upsampled too much on a fairly average CPU this
// difference is hardly noticeable and the results are far more pleasing to look at.
//
// This particular bicubic algorithm does pixel weighting according to a B-Spline that basically implements a Gaussian bell-like
// weighting kernel. Because of this method the results may appear a bit blurry when upsampling by large factors. This is basically
// because a slight gaussian blur is being performed to get the smooth look of the upsampled image.
// Edge pixels: 3-4 possible solutions
// - (Wrap/tile) Wrap the image, take the color value from the opposite side of the image.
// - (Mirror) Duplicate edge pixels, so that pixel at coordinate (2, n), where n is nonpositive, will have the value of (2, 1).
// - (Ignore) Simply ignore the edge pixels and apply the kernel only to pixels which do have all neighbours.
// - (Clamp) Choose the nearest pixel along the border. This takes the border pixels and extends them out to infinity.
//
// NOTE: below the y_offset and x_offset variables are being set for edge pixels using the "Mirror" method mentioned above
wxImage ret_image;
ret_image.Create(width, height, false);
unsigned char* src_data = M_IMGDATA->m_data;
unsigned char* src_alpha = M_IMGDATA->m_alpha;
unsigned char* dst_data = ret_image.GetData();
unsigned char* dst_alpha = NULL;
if(src_alpha)
{
ret_image.SetAlpha();
dst_alpha = ret_image.GetAlpha();
}
int k, i;
double srcpixx, srcpixy, dx, dy;
int dstx, dsty;
double sum_r = 0, sum_g = 0, sum_b = 0, sum_a = 0; // Sums for each color channel
int x_offset = 0, y_offset = 0;
double pixel_weight;
long src_pixel_index;
for(dsty = 0; dsty < height; dsty++)
{
// We need to calculate the source pixel to interpolate from - Y-axis
srcpixy = double(dsty) * M_IMGDATA->m_height / height;
dy = srcpixy - (int)srcpixy;
for(dstx = 0; dstx < width; dstx++)
{
// X-axis of pixel to interpolate from
srcpixx = double(dstx) * M_IMGDATA->m_width / width;
dx = srcpixx - (int)srcpixx;
// Clear all the RGBA sum values
sum_r = sum_g = sum_b = sum_a = 0;
// Here we actually determine the RGBA values for the destination pixel
for(k = -1; k <= 2; k++)
{
// Y offset
y_offset = srcpixy + double(k) < 0.0 ? 0 : (srcpixy + double(k) >= M_IMGDATA->m_height ? M_IMGDATA->m_height - 1 : srcpixy + k);
// Loop across the X axis
for(i = -1; i <= 2; i++)
{
// X offset
x_offset = srcpixx + double(i) < 0.0 ? 0 : (srcpixx + double(i) >= M_IMGDATA->m_width ? M_IMGDATA->m_width - 1 : srcpixx + i);
// Calculate the exact position where the source data should be pulled from based on the x_offset and y_offset
src_pixel_index = (y_offset * M_IMGDATA->m_width) + x_offset;
// Calculate the weight for the specified pixel according to the bicubic b-spline kernel we're using for interpolation
pixel_weight = spline_weight(double(i) - dx) * spline_weight(double(k) - dy);
// Create a sum of all velues for each color channel adjusted for the pixel's calculated weight
sum_r += double(src_data[src_pixel_index * 3 + 0]) * pixel_weight;
sum_g += double(src_data[src_pixel_index * 3 + 1]) * pixel_weight;
sum_b += double(src_data[src_pixel_index * 3 + 2]) * pixel_weight;
if(src_alpha)
sum_a += double(src_alpha[src_pixel_index]) * pixel_weight;
}
}
// Put the data into the destination image. The summed values are of double data type and are rounded here for accuracy
dst_data[0] = int(sum_r + 0.5);
dst_data[1] = int(sum_g + 0.5);
dst_data[2] = int(sum_b + 0.5);
dst_data += 3;
if(src_alpha)
*dst_alpha++ = sum_a;
}
}
return ret_image;
}
// Blur in the horizontal direction
wxImage wxImage::BlurHorizontal(int blurRadius)
{
wxImage ret_image;
ret_image.Create(M_IMGDATA->m_width, M_IMGDATA->m_height, false);
unsigned char* src_data = M_IMGDATA->m_data;
unsigned char* dst_data = ret_image.GetData();
unsigned char* src_alpha = M_IMGDATA->m_alpha;
unsigned char* dst_alpha = NULL;
// Check for a mask or alpha
if(M_IMGDATA->m_hasMask)
ret_image.SetMaskColour(M_IMGDATA->m_maskRed, M_IMGDATA->m_maskGreen, M_IMGDATA->m_maskBlue);
else
if(src_alpha)
{
ret_image.SetAlpha();
dst_alpha = ret_image.GetAlpha();
}
// Variables used in the blurring algorithm
int x, y;
int kernel_x;
long sum_r, sum_g, sum_b, sum_a;
long pixel_idx;
// Horizontal blurring algorithm - average all pixels in the specified blur radius in the X or horizontal direction
for(y = 0; y < M_IMGDATA->m_height; y++)
{
sum_r = sum_g = sum_b = sum_a = 0;
// Calculate the average of all pixels in the blur radius for the first pixel of the row
for(kernel_x = -blurRadius; kernel_x <= blurRadius; kernel_x++)
{
// To deal with the pixels at the start of a row so it's not grabbing GOK values from memory at negative indices of the image's data or grabbing from the previous row
if(kernel_x < 0)
pixel_idx = y * M_IMGDATA->m_width;
else
pixel_idx = kernel_x + y * M_IMGDATA->m_width;
sum_r += src_data[pixel_idx * 3 + 0];
sum_g += src_data[pixel_idx * 3 + 1];
sum_b += src_data[pixel_idx * 3 + 2];
sum_a += src_alpha ? src_alpha[pixel_idx] : 0;
}
dst_data[y * M_IMGDATA->m_width * 3 + 0] = sum_r / (blurRadius * 2 + 1);
dst_data[y * M_IMGDATA->m_width * 3 + 1] = sum_g / (blurRadius * 2 + 1);
dst_data[y * M_IMGDATA->m_width * 3 + 2] = sum_b / (blurRadius * 2 + 1);
if(src_alpha)
dst_alpha[y * M_IMGDATA->m_width] = sum_a / (blurRadius * 2 + 1);
// Now average the values of the rest of the pixels by just moving the blur radius box along the row
for(x = 1; x < M_IMGDATA->m_width; x++)
{
// Take care of edge pixels on the left edge by essentially duplicating the edge pixel
if(x - blurRadius - 1 < 0)
pixel_idx = y * M_IMGDATA->m_width;
else
pixel_idx = (x - blurRadius - 1) + y * M_IMGDATA->m_width;
// Subtract the value of the pixel at the left side of the blur radius box
sum_r -= src_data[pixel_idx * 3 + 0];
sum_g -= src_data[pixel_idx * 3 + 1];
sum_b -= src_data[pixel_idx * 3 + 2];
sum_a -= src_alpha ? src_alpha[pixel_idx] : 0;
// Take care of edge pixels on the right edge
if(x + blurRadius > M_IMGDATA->m_width - 1)
pixel_idx = M_IMGDATA->m_width - 1 + y * M_IMGDATA->m_width;
else
pixel_idx = x + blurRadius + y * M_IMGDATA->m_width;
// Add the value of the pixel being added to the end of our box
sum_r += src_data[pixel_idx * 3 + 0];
sum_g += src_data[pixel_idx * 3 + 1];
sum_b += src_data[pixel_idx * 3 + 2];
sum_a += src_alpha ? src_alpha[pixel_idx] : 0;
// Save off the averaged data
dst_data[x * 3 + y * M_IMGDATA->m_width * 3 + 0] = sum_r / (blurRadius * 2 + 1);
dst_data[x * 3 + y * M_IMGDATA->m_width * 3 + 1] = sum_g / (blurRadius * 2 + 1);
dst_data[x * 3 + y * M_IMGDATA->m_width * 3 + 2] = sum_b / (blurRadius * 2 + 1);
if(src_alpha)
dst_alpha[x + y * M_IMGDATA->m_width] = sum_a / (blurRadius * 2 + 1);
}
}
return ret_image;
}
// Blur in the vertical direction
wxImage wxImage::BlurVertical(int blurRadius)
{
wxImage ret_image;
ret_image.Create(M_IMGDATA->m_width, M_IMGDATA->m_height, false);
unsigned char* src_data = M_IMGDATA->m_data;
unsigned char* dst_data = ret_image.GetData();
unsigned char* src_alpha = M_IMGDATA->m_alpha;
unsigned char* dst_alpha = NULL;
// Check for a mask or alpha
if(M_IMGDATA->m_hasMask)
ret_image.SetMaskColour(M_IMGDATA->m_maskRed, M_IMGDATA->m_maskGreen, M_IMGDATA->m_maskBlue);
else
if(src_alpha)
{
ret_image.SetAlpha();
dst_alpha = ret_image.GetAlpha();
}
// Variables used in the blurring algorithm
int x, y;
int kernel_y;
long sum_r, sum_g, sum_b, sum_a;
long pixel_idx;
// Vertical blurring algorithm - same as horizontal but switched the opposite direction
for(x = 0; x < M_IMGDATA->m_width; x++)
{
sum_r = sum_g = sum_b = sum_a = 0;
// Calculate the average of all pixels in our blur radius box for the first pixel of the column
for(kernel_y = -blurRadius; kernel_y <= blurRadius; kernel_y++)
{
// To deal with the pixels at the start of a column so it's not grabbing GOK values from memory at negative indices of the image's data or grabbing from the previous column
if(kernel_y < 0)
pixel_idx = x;
else
pixel_idx = x + kernel_y * M_IMGDATA->m_width;
sum_r += src_data[pixel_idx * 3 + 0];
sum_g += src_data[pixel_idx * 3 + 1];
sum_b += src_data[pixel_idx * 3 + 2];
sum_a += src_alpha ? src_alpha[pixel_idx] : 0;
}
dst_data[x * 3 + 0] = sum_r / (blurRadius * 2 + 1);
dst_data[x * 3 + 1] = sum_g / (blurRadius * 2 + 1);
dst_data[x * 3 + 2] = sum_b / (blurRadius * 2 + 1);
if(src_alpha)
dst_alpha[x] = sum_a / (blurRadius * 2 + 1);
// Now average the values of the rest of the pixels by just moving the box along the column from top to bottom
for(y = 1; y < M_IMGDATA->m_height; y++)
{
// Take care of pixels that would be beyond the top edge by duplicating the top edge pixel for the column
if(y - blurRadius - 1 < 0)
pixel_idx = x;
else
pixel_idx = x + (y - blurRadius - 1) * M_IMGDATA->m_width;
// Subtract the value of the pixel at the top of our blur radius box
sum_r -= src_data[pixel_idx * 3 + 0];
sum_g -= src_data[pixel_idx * 3 + 1];
sum_b -= src_data[pixel_idx * 3 + 2];
sum_a -= src_alpha ? src_alpha[pixel_idx] : 0;
// Take care of the pixels that would be beyond the bottom edge of the image similar to the top edge
if(y + blurRadius > M_IMGDATA->m_height - 1)
pixel_idx = x + (M_IMGDATA->m_height - 1) * M_IMGDATA->m_width;
else
pixel_idx = x + (blurRadius + y) * M_IMGDATA->m_width;
// Add the value of the pixel being added to the end of our box
sum_r += src_data[pixel_idx * 3 + 0];
sum_g += src_data[pixel_idx * 3 + 1];
sum_b += src_data[pixel_idx * 3 + 2];
sum_a += src_alpha ? src_alpha[pixel_idx] : 0;
// Save off the averaged data
dst_data[(x + y * M_IMGDATA->m_width) * 3 + 0] = sum_r / (blurRadius * 2 + 1);
dst_data[(x + y * M_IMGDATA->m_width) * 3 + 1] = sum_g / (blurRadius * 2 + 1);
dst_data[(x + y * M_IMGDATA->m_width) * 3 + 2] = sum_b / (blurRadius * 2 + 1);
if(src_alpha)
dst_alpha[x + y * M_IMGDATA->m_width] = sum_a / (blurRadius * 2 + 1);
}
}
return ret_image;
}
// The new blur function
wxImage wxImage::Blur(int blurRadius)
{
wxImage ret_image;
ret_image.Create(M_IMGDATA->m_width, M_IMGDATA->m_height, false);
// Blur the image in each direction
ret_image = BlurHorizontal(blurRadius);
ret_image = ret_image.BlurVertical(blurRadius);
return ret_image;
}
wxImage wxImage::Rotate90( bool clockwise ) const
{
wxImage image;