blob: 9185d518fba60f8897f9ecc7a3a24f10f7a6ab5a [file] [log] [blame]
/*- pngpixel
* COPYRIGHT: Written by John Cunningham Bowler, 2011.
* To the extent possible under law, the author has waived all copyright and
* related or neighboring rights to this work. This work is published from:
* United States.
* Read a single pixel value from a PNG file.
* This code illustrates basic 'by-row' reading of a PNG file using libpng.
* Rows are read until a particular pixel is found; the value of this pixel is
* then printed on stdout.
* The code illustrates how to do this on interlaced as well as non-interlaced
* images. Normally you would call png_set_interlace_handling() to have libpng
* deal with the interlace for you, but that obliges you to buffer half of the
* image to assemble the interlaced rows. In this code
* png_set_interlace_handling() is not called and, instead, the code handles the
* interlace passes directly looking for the required pixel.
#include <stdlib.h>
#include <stdio.h>
#include <setjmp.h> /* required for error handling */
/* Normally use <png.h> here to get the installed libpng, but this is done to
* ensure the code picks up the local libpng implementation:
#include "../../png.h"
/* Return component 'c' of pixel 'x' from the given row. */
static unsigned int
component(png_const_bytep row, png_uint_32 x, unsigned int c,
unsigned int bit_depth, unsigned int channels)
/* PNG images can be up to 2^31 pixels wide, but this means they can be up to
* 2^37 bits wide (for a 64-bit pixel - the largest possible) and hence 2^34
* bytes wide. Since the row fitted into memory, however, the following must
* work:
png_uint_32 bit_offset_hi = bit_depth * ((x >> 6) * channels);
png_uint_32 bit_offset_lo = bit_depth * ((x & 0x3f) * channels + c);
row = (png_const_bytep)(((const png_byte (*)[8])row) + bit_offset_hi);
row += bit_offset_lo >> 3;
bit_offset_lo &= 0x07;
/* PNG pixels are packed into bytes to put the first pixel in the highest
* bits of the byte and into two bytes for 16-bit values with the high 8 bits
* first, so:
switch (bit_depth)
case 1: return (row[0] >> (7-bit_offset_lo)) & 0x01;
case 2: return (row[0] >> (6-bit_offset_lo)) & 0x03;
case 4: return (row[0] >> (4-bit_offset_lo)) & 0x0f;
case 8: return row[0];
case 16: return (row[0] << 8) + row[1];
/* This should never happen; it indicates a bug in this program or in
* libpng itself:
fprintf(stderr, "pngpixel: invalid bit depth %u\n", bit_depth);
/* Print a pixel from a row returned by libpng; determine the row format, find
* the pixel, and print the relevant information to stdout.
static void
print_pixel(png_structp png_ptr, png_infop info_ptr, png_const_bytep row,
png_uint_32 x)
unsigned int bit_depth = png_get_bit_depth(png_ptr, info_ptr);
switch (png_get_color_type(png_ptr, info_ptr))
printf("GRAY %u\n", component(row, x, 0, bit_depth, 1));
/* The palette case is slightly more difficult - the palette and, if
* present, the tRNS ('transparency', though the values are really
* opacity) data must be read to give the full picture:
int index = component(row, x, 0, bit_depth, 1);
png_colorp palette = NULL;
int num_palette = 0;
if ((png_get_PLTE(png_ptr, info_ptr, &palette, &num_palette) &
PNG_INFO_PLTE) && num_palette > 0 && palette != NULL)
png_bytep trans_alpha = NULL;
int num_trans = 0;
if ((png_get_tRNS(png_ptr, info_ptr, &trans_alpha, &num_trans,
NULL) & PNG_INFO_tRNS) && num_trans > 0 &&
trans_alpha != NULL)
printf("INDEXED %u = %d %d %d %d\n", index,
palette[index].red, palette[index].green,
index < num_trans ? trans_alpha[index] : 255);
else /* no transparency */
printf("INDEXED %u = %d %d %d\n", index,
palette[index].red, palette[index].green,
printf("INDEXED %u = invalid index\n", index);
printf("RGB %u %u %u\n", component(row, x, 0, bit_depth, 3),
component(row, x, 1, bit_depth, 3),
component(row, x, 2, bit_depth, 3));
printf("GRAY+ALPHA %u %u\n", component(row, x, 0, bit_depth, 2),
component(row, x, 1, bit_depth, 2));
printf("RGBA %u %u %u %u\n", component(row, x, 0, bit_depth, 4),
component(row, x, 1, bit_depth, 4),
component(row, x, 2, bit_depth, 4),
component(row, x, 3, bit_depth, 4));
png_error(png_ptr, "pngpixel: invalid color type");
int main(int argc, const char **argv)
/* This program uses the default, <setjmp.h> based, libpng error handling
* mechanism, therefore any local variable that exists before the call to
* setjmp and is changed after the call to setjmp returns successfully must
* be declared with 'volatile' to ensure that their values don't get
* destroyed by longjmp:
volatile int result = 1/*fail*/;
if (argc == 4)
long x = atol(argv[1]);
long y = atol(argv[2]);
FILE *f = fopen(argv[3], "rb");
volatile png_bytep row = NULL;
if (f != NULL)
/* libpng requires a callback function for handling errors; this
* callback must not return. The default callback function uses a
* stored <setjmp.h> style jmp_buf which is held in a png_struct and
* writes error messages to stderr. Creating the png_struct is a
* little tricky; just copy the following code.
png_structp png_ptr = png_create_read_struct(PNG_LIBPNG_VER_STRING,
if (png_ptr != NULL)
png_infop info_ptr = png_create_info_struct(png_ptr);
if (info_ptr != NULL)
/* Declare stack variables to hold pointers to locally allocated
* data.
/* Initialize the error control buffer: */
if (setjmp(png_jmpbuf(png_ptr)) == 0)
png_uint_32 width, height;
int bit_depth, color_type, interlace_method,
compression_method, filter_method;
png_bytep row_tmp;
/* Now associate the recently opened (FILE*) with the default
* libpng initialization functions. Sometimes libpng is
* compiled without stdio support (it can be difficult to do
* in some environments); in that case you will have to write
* your own read callback to read data from the (FILE*).
png_init_io(png_ptr, f);
/* And read the first part of the PNG file - the header and
* all the information up to the first pixel.
png_read_info(png_ptr, info_ptr);
/* This fills in enough information to tell us the width of
* each row in bytes, allocate the appropriate amount of
* space. In this case png_malloc is used - it will not
* return if memory isn't available.
row = png_malloc(png_ptr, png_get_rowbytes(png_ptr,
/* To avoid the overhead of using a volatile auto copy row_tmp
* to a local here - just use row for the png_free below.
row_tmp = row;
/* All the information we need is in the header is returned by
* png_get_IHDR, if this fails we can now use 'png_error' to
* signal the error and return control to the setjmp above.
if (png_get_IHDR(png_ptr, info_ptr, &width, &height,
&bit_depth, &color_type, &interlace_method,
&compression_method, &filter_method))
int passes, pass;
/* png_set_interlace_handling returns the number of
* passes required as well as turning on libpng's
* handling, but since we do it ourselves this is
* necessary:
switch (interlace_method)
passes = 1;
png_error(png_ptr, "pngpixel: unknown interlace");
/* Now read the pixels, pass-by-pass, row-by-row: */
for (pass=0; pass<passes; ++pass)
png_uint_32 ystart, xstart, ystep, xstep;
png_uint_32 py;
if (interlace_method == PNG_INTERLACE_ADAM7)
/* Sometimes the whole pass is empty because the
* image is too narrow or too short. libpng
* expects to be called for each row that is
* present in the pass, so it may be necessary to
* skip the loop below (over py) if the image is
* too narrow.
if (PNG_PASS_COLS(width, pass) == 0)
/* We need the starting pixel and the offset
* between each pixel in this pass; use the macros
* in png.h:
xstart = PNG_PASS_START_COL(pass);
ystart = PNG_PASS_START_ROW(pass);
xstep = PNG_PASS_COL_OFFSET(pass);
ystep = PNG_PASS_ROW_OFFSET(pass);
ystart = xstart = 0;
ystep = xstep = 1;
/* To find the pixel, loop over 'py' for each pass
* reading a row and then checking to see if it
* contains the pixel.
for (py = ystart; py < height; py += ystep)
png_uint_32 px, ppx;
/* png_read_row takes two pointers. When libpng
* handles the interlace the first is filled in
* pixel-by-pixel, and the second receives the same
* pixels but they are replicated across the
* unwritten pixels so far for each pass. When we
* do the interlace, however, they just contain
* the pixels from the interlace pass - giving
* both is wasteful and pointless, so we pass a
* NULL pointer.
png_read_row(png_ptr, row_tmp, NULL);
/* Now find the pixel if it is in this row; there
* are, of course, much better ways of doing this
* than using a for loop:
if (y == py) for (px = xstart, ppx = 0;
px < width; px += xstep, ++ppx) if (x == px)
/* 'ppx' is the index of the pixel in the row
* buffer.
print_pixel(png_ptr, info_ptr, row_tmp, ppx);
/* Now terminate the loops early - we have
* found and handled the required data.
goto pass_loop_end;
} /* x loop */
} /* y loop */
} /* pass loop */
/* Finally free the temporary buffer: */
row = NULL;
png_free(png_ptr, row_tmp);
png_error(png_ptr, "pngpixel: png_get_IHDR failed");
/* Else libpng has raised an error. An error message has
* already been output, so it is only necessary to clean up
* locally allocated data:
if (row != NULL)
/* The default implementation of png_free never errors out
* (it just crashes if something goes wrong), but the safe
* way of using it is still to clear 'row' before calling
* png_free:
png_bytep row_tmp = row;
row = NULL;
png_free(png_ptr, row_tmp);
png_destroy_info_struct(png_ptr, &info_ptr);
fprintf(stderr, "pngpixel: out of memory allocating png_info\n");
png_destroy_read_struct(&png_ptr, NULL, NULL);
fprintf(stderr, "pngpixel: out of memory allocating png_struct\n");
fprintf(stderr, "pngpixel: %s: could not open file\n", argv[3]);
/* Wrong number of arguments */
fprintf(stderr, "pngpixel: usage: pngpixel x y png-file\n");
return result;
#endif /* READ && SEQUENTIAL_READ */