Søren,
Thank you for applying the patch.
I have look at the pngread.cc a little bit more and find out that
the reason for wrong png to matrix conversion was essentially in the
load_canvas function of this source file.
There is a lot of conversion of the input png file in the load_canvas function:
it always convertes to RGB (even if it is gray scale), alpha channel
is always added if missing. But png info structure is never updated to
reflect these changes.
I attached the patch which fixes these issues: it keep track of
bit_depth and color_map changes and updates png info structure
accordingly. As result when load_canvas return the main function
"pngread" will have proper values.
This patch also support 16 bits per color images, original code
converts everything to 8bits per color, which is bad idea for
scientific data. I often work with small signal on a large background
so reducing accuracy is unacceptable. I commented out 16 to 8 bits
conversion code.
Accordingly I added code which is capable of reading 16bit per color raw data.
General notes: looks like the pngwrite should be cleaned up a lot, it
process cases which will never occur, for example bit_depth<8. The
load_canvas converts everything to at least 8 bit. Gray scale images
never arrive to the pngwrite for similar reasons.
On Fri, Mar 13, 2009 at 5:34 AM, Søren Hauberg <[email protected]> wrote:
> tor, 12 03 2009 kl. 21:15 -0400, skrev Eugeniy Mikhailov:
>> I am attaching the patch fixing this issue.
>
> I've applied this patch.
--
Eugeniy
--- /tmp/image/src/pngread.cc 2009-03-13 05:33:49.000000000 -0400
+++ /tmp/image.new_by_evmik/src/pngread.cc 2009-03-19 13:19:53.000000000 -0400
@@ -84,13 +84,14 @@
((pic->color_type == PNG_COLOR_TYPE_PALETTE) && (pic->bit_depth == 1)) )
dim(2) = 1;
- if (pic->bit_depth > 1 && pic->bit_depth < 8)
- pic->bit_depth = 8;
+ if (pic->bit_depth > 1 && pic->bit_depth < 8)
+ pic->bit_depth = 8; //this should never happen according to load_canvas code
int isAlfaChannelPresent = 0;
if ( pic->color_type & PNG_COLOR_MASK_ALPHA)
isAlfaChannelPresent=1;
+
NDArray out(dim);
dim.resize(2);
@@ -98,6 +99,8 @@
Array<int> coord = Array<int> (3);
+ int major_byte, minor_byte,row_pxl_position;
+ /* calculate the number of color channels (including alpha) per pixel */
int ElementsPerPixel=out.dims()(2)+isAlfaChannelPresent;
for (unsigned long j=0; j < pic->height; j++) {
coord(0) = j;
@@ -106,13 +109,42 @@
for (int c = 0; c < out.dims()(2); c++) {
coord(2) = c;
- out(coord) = pic->row_pointers[j][i*ElementsPerPixel+c];
+ switch(pic->bit_depth) {
+ case 8:
+ row_pxl_position=(i*ElementsPerPixel+c);
+ out(coord) = pic->row_pointers[j][row_pxl_position];
+ break;
+ case 16:
+ // converting big endian
+ row_pxl_position=2*(i*ElementsPerPixel+c);
+ major_byte=pic->row_pointers[j][row_pxl_position];
+ minor_byte=pic->row_pointers[j][row_pxl_position+1] ;
+ out(coord) = major_byte*256+minor_byte;
+ break;
+ default:
+ printf("do not know how to handle bit depth of %d\n",pic->bit_depth);
+ }
+ }
+ if (isAlfaChannelPresent) { // it always should according to load canvas code
+ switch(pic->bit_depth) {
+ case 8:
+ row_pxl_position=(i*ElementsPerPixel+ElementsPerPixel-1);
+ alpha(j,i) = pic->row_pointers[j][row_pxl_position];
+ break;
+ case 16:
+ // converting big endian
+ row_pxl_position=2*(i*ElementsPerPixel+ElementsPerPixel-1);
+ major_byte = pic->row_pointers[j][row_pxl_position];
+ minor_byte = pic->row_pointers[j][row_pxl_position+1];
+ alpha(j,i) = major_byte*256+minor_byte;
+ break;
+ default:
+ printf("do not know how to handle bit depth of %d\n",pic->bit_depth);
+ }
+ } else {
+ alpha(j,i) = 255;
+ }
}
- if (isAlfaChannelPresent)
- alpha(j,i) = pic->row_pointers[j][i*ElementsPerPixel+(ElementsPerPixel-1)];
- else
- alpha(j,i) = 1;
- }
}
out = out.squeeze();
@@ -197,7 +229,9 @@
png_set_palette_to_rgb(png_ptr);
}
if (color_type == PNG_COLOR_TYPE_GRAY && bit_depth < 8) {
- png_set_gray_1_2_4_to_8(png_ptr);
+ png_set_gray_1_2_4_to_8(png_ptr); // this function deprecated need to be redone
+ bit_depth=8;
+ info_ptr->bit_depth=bit_depth;
}
if (png_get_valid(png_ptr, info_ptr, PNG_INFO_tRNS)) { //add alpha
png_set_tRNS_to_alpha(png_ptr);
@@ -206,21 +240,35 @@
// Always transform image to RGB
if (color_type == PNG_COLOR_TYPE_GRAY
|| color_type == PNG_COLOR_TYPE_GRAY_ALPHA)
+ {
png_set_gray_to_rgb(png_ptr);
+ color_type= (color_type | PNG_COLOR_MASK_COLOR);
+ info_ptr->color_type=color_type;
+ }
// If no alpha layer is present, create one
- if (color_type == PNG_COLOR_TYPE_GRAY
- || color_type == PNG_COLOR_TYPE_RGB)
- png_set_add_alpha(png_ptr, 0xff, PNG_FILLER_AFTER);
+ if (!(color_type & PNG_COLOR_MASK_ALPHA))
+ {
+ png_set_add_alpha(png_ptr, 0xff, PNG_FILLER_AFTER);
+ color_type= (color_type | PNG_COLOR_MASK_ALPHA);
+ info_ptr->color_type=color_type;
+ }
if (bit_depth < 8) {
png_set_packing(png_ptr);
+ bit_depth=8;
+ info_ptr->bit_depth=bit_depth;
}
-
+
+ // Hey! Our signal could be small and in the lower bits,
+ // leave our data alone and do not decrease accuracy
+ // 16 -> 8 bits commented out
// For now, use 8-bit only
- if (bit_depth == 16) {
- png_set_strip_16(png_ptr);
- }
+ //if (bit_depth == 16) {
+ //png_set_strip_16(png_ptr);
+ //bit_depth=8;
+ //info_ptr->bit_depth=bit_depth;
+ //}
png_read_update_info(png_ptr,info_ptr);
------------------------------------------------------------------------------
Apps built with the Adobe(R) Flex(R) framework and Flex Builder(TM) are
powering Web 2.0 with engaging, cross-platform capabilities. Quickly and
easily build your RIAs with Flex Builder, the Eclipse(TM)based development
software that enables intelligent coding and step-through debugging.
Download the free 60 day trial. http://p.sf.net/sfu/www-adobe-com
_______________________________________________
Octave-dev mailing list
[email protected]
https://lists.sourceforge.net/lists/listinfo/octave-dev
