ImageMagick colors depth conversion - imagemagick

Question

I have several thousands of tga files (without palette) which contain RGBA4444 data (I know usualy tga files don't contain RGBA4444 data). I would like to convert them into RGBA8888 data. I use the following command line:
convert -depth 4 woody4.tga -depth 8 woody8.tga
In this case, woody4.tga is the original RGBA4444 file, and woody8.tga the target RGBA8888 file but it doesn't change the colors of my pictures, what am I missing?
Thanks,
Pierre
Edit:
Thanks very much Mark, I have successfully converted more than 10 000 TGA with your program, the result is very good and correct to the original TGA ! this would has been impossible without the parallel command ! Just a last point, I have around 50 TGA larger (the backgrounds of the game) which are coded with RGBA5650 and not RGBA4444, how can I modify your program to manage the RGBA5650 ? Thanks very much !

Answer 1

Oh, I see Eric beat me to it:-)
Hey ho! I did it a different way anyway and got a different answer so you can see which one you like best. I also wrote some C but I didn't rely on any libraries, I just read the TGA and converted it to a PAM format and let ImageMagick make that into PNG afterwards at command-line.
I chose PAM because it is the simplest file to write which supports transparency - see Wikipedia on PAM format.
#include <stdio.h>
#include <stdlib.h>
#include <fcntl.h>
#include <sys/types.h>
#include <sys/stat.h>
int main(int argc,char* argv[]){
unsigned char buf[64];
FILE* fp=fopen(argv[1],"rb");
if(fp==NULL){
fprintf(stderr,"ERROR: Unable to open %s\n",argv[1]);
exit(1);
}
// Read TGA header of 18 bytes, extract width and height
fread(buf,1,18,fp); // 12 bytes junk, 2 bytes width, 2 bytes height, 2 bytes junk
unsigned short w=buf[12]|(buf[13]<<8);
unsigned short h=buf[14]|(buf[15]<<8);
// Write PAM header
fprintf(stdout,"P7\n");
fprintf(stdout,"WIDTH %d\n",w);
fprintf(stdout,"HEIGHT %d\n",h);
fprintf(stdout,"DEPTH 4\n");
fprintf(stdout,"MAXVAL 255\n");
fprintf(stdout,"TUPLTYPE RGB_ALPHA\n");
fprintf(stdout,"ENDHDR\n");
// Read 2 bytes at a time RGBA4444
while(fread(buf,2,1,fp)==1){
unsigned char out[4];
out[0]=(buf[1]&0x0f)<<4;
out[1]=buf[0]&0xf0;
out[2]=(buf[0]&0x0f)<<4;
out[3]=buf[1]&0xf0;
// Write the 4 modified bytes out RGBA8888
fwrite(out,4,1,stdout);
}
fclose(fp);
return 0;
}
I the compile that with gcc:
gcc targa.c -o targa
Or you could use clang:
clang targa.c -o targa
and run it with
./targa someImage.tga > someImage.pam
and convert the PAM to PNG with ImageMagick at the command-line:
convert someImage.pam someImage.png
If you want to avoid writing the intermediate PAM file to disk, you can pipe it straight into convert like this:
./targa illu_evolution_01.tga | convert - result.png
You can, equally, make a BMP output file if you wish:
./targa illu_evolution_01.tga | convert - result.bmp
If you have thousands of files to do, and you are on a Mac or Linux, you can use GNU Parallel and get them all done in parallel much faster like this:
parallel --eta './targa {} | convert - {.}.png' ::: *.tga
If you have more than a couple of thousand files, you may get "Argument list too long" errors, in which case, use the slightly harder syntax:
find . -name \*tga -print0 | parallel -0 --eta './targa {} | convert - {.}.png'
On a Mac, you would install GNU Parallel with homebrew using:
brew install parallel
For your RGBA5650 images, I will fall back to PPM as my intermediate format because the alpha channel of PAM is no longer needed. The code will now look like this:
#include <stdio.h>
#include <stdlib.h>
#include <fcntl.h>
#include <sys/types.h>
#include <sys/stat.h>
int main(int argc,char* argv[]){
unsigned char buf[64];
FILE* fp=fopen(argv[1],"rb");
if(fp==NULL){
fprintf(stderr,"ERROR: Unable to open %s\n",argv[1]);
exit(1);
}
// Read TGA header of 18 bytes, extract width and height
fread(buf,1,18,fp); // 12 bytes junk, 2 bytes width, 2 bytes height, 2 bytes junk
unsigned short w=buf[12]|(buf[13]<<8);
unsigned short h=buf[14]|(buf[15]<<8);
// Write PPM header
fprintf(stdout,"P6\n");
fprintf(stdout,"%d %d\n",w,h);
fprintf(stdout,"255\n");
// Read 2 bytes at a time RGBA5650
while(fread(buf,2,1,fp)==1){
unsigned char out[3];
out[0]=buf[1]&0xf8;
out[1]=((buf[1]&7)<<5) | ((buf[0]>>3)&0x1c);
out[2]=(buf[0]&0x1f)<<3;
// Write the 3 modified bytes out RGB888
fwrite(out,3,1,stdout);
}
fclose(fp);
return 0;
}
And will compile and run exactly the same way.

Answer 2

Updated answer.
After reading a few documents about TARGA format. I've revised + simplified a C program to convert.
// tga2img.c
#include <stdio.h>
#include <stdlib.h>
#include <wand/MagickWand.h>
typedef struct {
unsigned char idlength;
unsigned char colourmaptype;
unsigned char datatypecode;
short int colourmaporigin;
short int colourmaplength;
unsigned char colourmapdepth;
short int x_origin;
short int y_origin;
short int width;
short int height;
unsigned char bitsperpixel;
unsigned char imagedescriptor;
} HEADER;
typedef struct {
int extensionoffset;
int developeroffset;
char signature[16];
unsigned char p;
unsigned char n;
} FOOTER;
int main(int argc, const char * argv[]) {
HEADER tga_header;
FOOTER tga_footer;
FILE
* fd;
size_t
tga_data_size,
tga_pixel_size,
i,
j;
unsigned char
* tga_data,
* buffer;
const char
* input,
* output;
if (argc != 3) {
printf("Usage:\n\t %s <input> <output>\n", argv[0]);
return 1;
}
input = argv[1];
output = argv[2];
fd = fopen(input, "rb");
if (fd == NULL) {
fprintf(stderr, "Unable to read TGA input\n");
return 1;
}
/********\
* TARGA *
\*********/
#pragma mark TARGA
// Read TGA header
fread(&tga_header.idlength, sizeof(unsigned char), 1, fd);
fread(&tga_header.colourmaptype, sizeof(unsigned char), 1, fd);
fread(&tga_header.datatypecode, sizeof(unsigned char), 1, fd);
fread(&tga_header.colourmaporigin, sizeof( short int), 1, fd);
fread(&tga_header.colourmaplength, sizeof( short int), 1, fd);
fread(&tga_header.colourmapdepth, sizeof(unsigned char), 1, fd);
fread(&tga_header.x_origin, sizeof( short int), 1, fd);
fread(&tga_header.y_origin, sizeof( short int), 1, fd);
fread(&tga_header.width, sizeof( short int), 1, fd);
fread(&tga_header.height, sizeof( short int), 1, fd);
fread(&tga_header.bitsperpixel, sizeof(unsigned char), 1, fd);
fread(&tga_header.imagedescriptor, sizeof(unsigned char), 1, fd);
// Calculate sizes
tga_pixel_size = tga_header.bitsperpixel / 8;
tga_data_size = tga_header.width * tga_header.height * tga_pixel_size;
// Read image data
tga_data = malloc(tga_data_size);
fread(tga_data, 1, tga_data_size, fd);
// Read TGA footer.
fseek(fd, -26, SEEK_END);
fread(&tga_footer.extensionoffset, sizeof( int), 1, fd);
fread(&tga_footer.developeroffset, sizeof( int), 1, fd);
fread(&tga_footer.signature, sizeof( char), 16, fd);
fread(&tga_footer.p, sizeof(unsigned char), 1, fd);
fread(&tga_footer.n, sizeof(unsigned char), 1, fd);
fclose(fd);
buffer = malloc(tga_header.width * tga_header.height * 4);
#pragma mark RGBA4444 to RGBA8888
for (i = 0, j=0; i < tga_data_size; i+= tga_pixel_size) {
buffer[j++] = (tga_data[i+1] & 0x0f) << 4; // Red
buffer[j++] = tga_data[i ] & 0xf0; // Green
buffer[j++] = (tga_data[i ] & 0x0f) << 4; // Blue
buffer[j++] = tga_data[i+1] & 0xf0; // Alpha
}
free(tga_data);
/***************\
* IMAGEMAGICK *
\***************/
#pragma mark IMAGEMAGICK
MagickWandGenesis();
PixelWand * background;
background = NewPixelWand();
PixelSetColor(background, "none");
MagickWand * wand;
wand = NewMagickWand();
MagickNewImage(wand,
tga_header.width,
tga_header.height,
background);
background = DestroyPixelWand(background);
MagickImportImagePixels(wand,
0,
0,
tga_header.width,
tga_header.height,
"RGBA",
CharPixel,
buffer);
free(buffer);
MagickWriteImage(wand, argv[2]);
wand = DestroyMagickWand(wand);
return 0;
}
Which can be compiled with clang $(MagickWand-config --cflags --libs) -o tga2im tga2im.c, and can be executed simply by ./tga2im N_birthday_0000.tga N_birthday_0000.tga.png.
Original answer.
The only way I can think of converting the images is to author a quick program/script to do the bitwise color-pixel logic.
This answer offers a quick way to read the image data; so combining with MagickWand, can be converted easily. (Although I know there'll be better solutions found on old game-dev forums...)
#include <stdio.h>
#include <stdbool.h>
#include <wand/MagickWand.h>
typedef struct
{
unsigned char imageTypeCode;
short int imageWidth;
short int imageHeight;
unsigned char bitCount;
unsigned char *imageData;
} TGAFILE;
bool LoadTGAFile(const char *filename, TGAFILE *tgaFile);
int main(int argc, const char * argv[]) {
const char
* input,
* output;
if (argc != 3) {
printf("Usage:\n\t%s <input> <output>\n", argv[0]);
}
input = argv[1];
output = argv[2];
MagickWandGenesis();
TGAFILE header;
if (LoadTGAFile(input, &header) == true) {
// Build a blank canvas image matching TGA file.
MagickWand * wand;
wand = NewMagickWand();
PixelWand * background;
background = NewPixelWand();
PixelSetColor(background, "NONE");
MagickNewImage(wand, header.imageWidth, header.imageHeight, background);
background = DestroyPixelWand(background);
// Allocate RGBA8888 buffer
unsigned char * buffer = malloc(header.imageWidth * header.imageHeight * 4);
// Iterate over TGA image data, and convert RGBA4444 to RGBA8888;
size_t pixel_size = header.bitCount / 8;
size_t total_bytes = header.imageWidth * header.imageHeight * pixel_size;
for (int i = 0, j = 0; i < total_bytes; i+=pixel_size) {
// Red
buffer[j++] = (header.imageData[i ] & 0x0f) << 4;
// Green
buffer[j++] = (header.imageData[i ] & 0xf0);
// Blue
buffer[j++] = (header.imageData[i+1] & 0xf0) << 4;
// Alpha
buffer[j++] = (header.imageData[i+1] & 0xf0);
}
// Import image data over blank canvas
MagickImportImagePixels(wand, 0, 0, header.imageWidth, header.imageHeight, "RGBA", CharPixel, buffer);
// Write image
MagickWriteImage(wand, output);
wand = DestroyMagickWand(wand);
} else {
fprintf(stderr, "Could not read TGA file %s\n", input);
}
MagickWandTerminus();
return 0;
}
/*
* Method copied verbatim from https://stackoverflow.com/a/7050007/438117
* Show your love by +1 to Wroclai answer.
*/
bool LoadTGAFile(const char *filename, TGAFILE *tgaFile)
{
FILE *filePtr;
unsigned char ucharBad;
short int sintBad;
long imageSize;
int colorMode;
unsigned char colorSwap;
// Open the TGA file.
filePtr = fopen(filename, "rb");
if (filePtr == NULL)
{
return false;
}
// Read the two first bytes we don't need.
fread(&ucharBad, sizeof(unsigned char), 1, filePtr);
fread(&ucharBad, sizeof(unsigned char), 1, filePtr);
// Which type of image gets stored in imageTypeCode.
fread(&tgaFile->imageTypeCode, sizeof(unsigned char), 1, filePtr);
// For our purposes, the type code should be 2 (uncompressed RGB image)
// or 3 (uncompressed black-and-white images).
if (tgaFile->imageTypeCode != 2 && tgaFile->imageTypeCode != 3)
{
fclose(filePtr);
return false;
}
// Read 13 bytes of data we don't need.
fread(&sintBad, sizeof(short int), 1, filePtr);
fread(&sintBad, sizeof(short int), 1, filePtr);
fread(&ucharBad, sizeof(unsigned char), 1, filePtr);
fread(&sintBad, sizeof(short int), 1, filePtr);
fread(&sintBad, sizeof(short int), 1, filePtr);
// Read the image's width and height.
fread(&tgaFile->imageWidth, sizeof(short int), 1, filePtr);
fread(&tgaFile->imageHeight, sizeof(short int), 1, filePtr);
// Read the bit depth.
fread(&tgaFile->bitCount, sizeof(unsigned char), 1, filePtr);
// Read one byte of data we don't need.
fread(&ucharBad, sizeof(unsigned char), 1, filePtr);
// Color mode -> 3 = BGR, 4 = BGRA.
colorMode = tgaFile->bitCount / 8;
imageSize = tgaFile->imageWidth * tgaFile->imageHeight * colorMode;
// Allocate memory for the image data.
tgaFile->imageData = (unsigned char*)malloc(sizeof(unsigned char)*imageSize);
// Read the image data.
fread(tgaFile->imageData, sizeof(unsigned char), imageSize, filePtr);
// Change from BGR to RGB so OpenGL can read the image data.
for (int imageIdx = 0; imageIdx < imageSize; imageIdx += colorMode)
{
colorSwap = tgaFile->imageData[imageIdx];
tgaFile->imageData[imageIdx] = tgaFile->imageData[imageIdx + 2];
tgaFile->imageData[imageIdx + 2] = colorSwap;
}
fclose(filePtr);
return true;
}
The order of the color channels may need to be switch around.

Answer 3

I have been thinking about this some more and it ought to be possible to reconstruct the image without any special software - I can't quite see my mistake for the moment by maybe #emcconville can cast your expert eye over it and point out my mistake! Pretty please?
So, my concept is that ImageMagick has read in the image size and pixel data correctly but has just allocated the bits according to the standard RGB5551 interpretation of a TARGA file rather than RGBA4444. So, we rebuild the 16-bits of data it read and split them differently.
The first line below does the rebuild into the original 16-bit data, then each subsequent line splits out one of the RGBA channels and then we recombine them:
convert illu_evolution_01.tga -depth 16 -channel R -fx "(((r*255)<<10) | ((g*255)<<5) | (b*255) | ((a*255)<<15))/255" \
\( -clone 0 -channel R -fx "((((r*255)>>12)&15)<<4)/255" \) \
\( -clone 0 -channel R -fx "((((r*255)>>8 )&15)<<4)/255" \) \
\( -clone 0 -channel R -fx "((((r*255) )&15)<<4)/255" \) \
-delete 0 -set colorspace RGB -combine -colorspace sRGB result.png
# The rest is just debug so you can see the reconstructed channels in [rgba].png
convert result.png -channel R -separate r.png
convert result.png -channel G -separate g.png
convert result.png -channel B -separate b.png
convert result.png -channel A -separate a.png
So, the following diagram represents the 16-bits of 1 pixel:
A R R R R R G G G G G B B B B B <--- what IM saw
R R R R G G G G B B B B A A A A <--- what it really meant
Yes, I have disregarded the alpha channel for the moment.