The SGI Image File FormatVersion 1.00
Paul Haeberli
IntroductionThis is the definitive document describing the SGI image file format. This is a low level spec that describes the actual byte level format of SGI image files. On SGI machines the preferred way of reading and writing SGI image files is to use the image library -limage. This library provides a set of functions that make it easy to read and write SGI images. If you are on an SGI workstation you can get info on -limage by doing: % man 4 rgb A note on byte order of values in the SGI image filesIn the following description a notation like bits[7..0] is used to denote a range of bits in a binary value. Bit 0 is the lowest order bit in the value. All short values are represented by 2 bytes. The first byte stores the high order 8 bits of the value: bits[15..8]. The second byte stores the low order 8 bits of the value: bits[7..0]. So this function will read a short value from the file: unsigned short getshort(inf) FILE *inf; { unsigned char buf[2]; fread(buf,2,1,inf); return (buf[0]<<8)+(buf[1]<<0); } All long values are represented by 4 bytes. The first byte stores the high order 8 bits of the value: bits[31..24]. The second byte stores bits[23..16]. The third byte stores bits[15..8]. The forth byte stores the low order 8 bits of the value: bits[7..0]. And this function will read a long value from the file: static long getlong(inf) FILE *inf; { unsigned char buf[4]; fread(buf,4,1,inf); return (buf[0]<<24)+(buf[1]<<16)+(buf[2]<<8)+(buf[3]<<0); }
The general structure of an SGI image fileThe header indicates whether the image is run length encoded (RLE).If the image is not run length encoded, this is the structure: The Header The Image DataIf the image is run length encoded, this is the structure: The Header The Offset Tables The Image Data The HeaderThe header consists of the following: Size | Type | Name | Description 2 bytes | short | MAGIC | IRIS image file magic number 1 byte | char | STORAGE | Storage format 1 byte | char | BPC | Number of bytes per pixel channel 2 bytes | ushort | DIMENSION | Number of dimensions 2 bytes | ushort | XSIZE | X size in pixels 2 bytes | ushort | YSIZE | Y size in pixels 2 bytes | ushort | ZSIZE | Number of channels 4 bytes | long | PIXMIN | Minimum pixel value 4 bytes | long | PIXMAX | Maximum pixel value 4 bytes | char | DUMMY | Ignored 80 bytes | char | IMAGENAME | Image name 4 bytes | long | COLORMAP | Colormap ID 404 bytes | char | DUMMY | Ignored Here is a description of each field in the image file header: MAGIC - This is the decimal value 474 saved as a short. This identifies the file as an SGI image file. STORAGE - specifies whether the image is stored using run length encoding (RLE) or not (VERBATIM). If RLE is used, the value of this byte will be 1. Otherwise the value of this byte will be 0. The only allowed values for this field are 0 or 1. BPC - describes the precision that is used to store each channel of an image. This is the number of bytes per pixel component. The majority of SGI image files use 1 byte per pixel component, giving 256 levels. Some SGI image files use 2 bytes per component. The only allowed values for this field are 1 or 2. DIMENSION - described the number of dimensions in the data stored in the image file. The only allowed values are 1, 2, or 3. If this value is 1, the image file consists of only 1 channel and only 1 scanline (row). The length of this scanline is given by the value of XSIZE below. If this value is 2, the file consists of a single channel with a number of scanlines. The width and height of the image are given by the values of XSIZE and YSIZE below. If this value is 3, the file consists of a number of channels. The width and height of the image are given by the values of XSIZE and YSIZE below. The number of channels is given by the value of ZSIZE below. XSIZE - The width of the image in pixels YSIZE - The height of the image in pixels ZSIZE - The number of channels in the image. B/W (greyscale) images are stored as 2 dimensional images with a ZSIZE or 1. RGB color images are stored as 3 dimensional images with a ZSIZE of 3. An RGB image with an ALPHA channel is stored as a 3 dimensional image with a ZSIZE of 4. There are no inherent limitations in the SGI image file format that would preclude the creation of image files with more than 4 channels. PINMIN - The minimum pixel value in the image. The value of 0 may be used if no pixel has a value that is smaller than 0. PINMAX - The maximum pixel value in the image. The value of 255 may be used if no pixel has a value that is greater than 255. This is the value that is considered to be full brightness in the image. DUMMY - This 4 bytes of data should be set to 0. IMAGENAME - An null terminated ascii string of up to 79 characters terminated by a null may be included here. This is not commonly used. COLORMAP - This controls how the pixel values in the file should be interpreted. It can have one of these four values: 0: NORMAL - The data in the channels represent B/W values for images with 1 channel, RGB values for images with 3 channels, and RGBA values for images with 4 channels. Almost all the SGI image files are of this type. 1: DITHERED - The image will have only 1 channel of data. For each pixel, RGB data is packed into one 8 bit value. 3 bits are used for red and green, while blue uses 2 bits. Red data is found in bits[2..0], green data in bits[5..3], and blue data in bits[7..6]. This format is obsolete. 2: SCREEN - The image will have only 1 channel of data. This format was used to store color-indexed pixels. To convert the pixel values into RGB values a colormap must be used. The appropriate color map varies from image to image. This format is obsolete. 3: COLORMAP - The image is used to store a color map from an SGI machine. In this case the image is not displayable in the conventional sense. DUMMY - This 404 bytes of data should be set to 0. This makes the header exactly 512 bytes.
The Image Data (if not RLE)If the image is stored verbatim (without RLE), then image data directly follows the 512 byte header. The data for each scanline of the first channel is written first. If the image has more than 1 channel, all the data for the first channel is written, followed by the remaining channels. If the BPC value is 1, then each scanline is written as XSIZE bytes. If the BPC value is 2, then each scanline is written as XSIZE shorts. These shorts are stored in the byte order described above.
The Offset Tables (if RLE)If the image is stored using run length encoding, offset tables follow the header that describe what the file offsets are to the RLE for each scanline. This information only applies if the value for STORAGE above is 1. Size | Type | Name | Description tablen longs | long | STARTTAB | Start table tablen longs | long | LENGTHTAB | Length table One entry in each table is needed for each scanline of RLE data. The total number of scanlines in the image (tablen) is determined by the product of the YSIZE and ZSIZE. There are two tables of longs that are written. Each consists of tablen longs of data. The first table has the file offsets to the RLE data for each scanline in the image. In a file with more than 1 channel (ZSIZE > 1) this table first has all the offsets for the scanlines in the first channel, followed be offsets for the scanlines in the second channel, etc. The second table has the RLE data length for each scanline in the image. In a file with more than 1 channel (ZSIZE > 1) this table first has all the RLE data lengths for the scanlines in the first channel, followed be RLE data lengths for the scanlines in the second channel, etc. To find the the file offset, and the number of bytes in the RLE data for a particular scanline, these two arrays may be read in and indexed as follows: To read in the tables: unsigned long *starttab, *lengthtab; tablen = YSIZE*ZSIZE*sizeof(long); starttab = (unsigned long *)mymalloc(tablen); lengthtab = (unsigned long *)mymalloc(tablen); fseek(inf,512,SEEK_SET); readlongtab(inf,starttab); readlongtab(ing,lengthtab); To find the file offset and RLE data length for a scanline: rowno is an integer in the range 0 to YSIZE-1 channo is an integer in the range 0 to ZSIZE-1 rleoffset = starttab[rowno+channo*YSIZE] rlelength = lengthtab[rowno+channo*YSIZE] It is possible for two identical rows (scanlines) to share compressed data. A completely white image could be written as a single compressed row and having all table entries point to that row. Another little hack that should work is if you are writing out a RGB RLE file, and a particular scanline is achromatic (greyscale), you could just make the r, g and b rows point to the same data!! The Image Data (if RLE)This information only applies if the value for STORAGE above is 1. If the image is stored using run length encoding, the image data follows the offset tables above. The RLE data is not in any particular order. The offset tables above are used to locate the rle data for any scanline. The RLE data must be read in from the file and expanded into pixel data in the following manner: If BPC is 1, then there is one byte per pixel. In this case the RLE data should be read into an array of chars. To expand data, the low order seven bits of the first byte: bits[6..0] are used to form a count. If the high order bit of the first byte is 1: bit[7], then the count is used to specify how many bytes to copy from the RLE data buffer to the destination. Otherwise, if the high order bit of the first byte is 0: bit[7], then the count is used to specify how many times to repeat the value of the following byte, in the destination. This process continues until a count of 0 is found. This should decompress exactly XSIZE pixels. Here is example code to decompress a scanline: expandrow(optr,iptr,z) unsigned char *optr, *iptr; int z; { unsigned char pixel, count; optr += z; while(1) { pixel = *iptr++; if ( !(count = (pixel & 0x7f)) ) return; if(pixel & 0x80) { while(count--) *optr++ = *iptr++; } else { pixel = *iptr++; while(count--) *optr++ = pixel; } } } If BPC is 2, there is one short (2 bytes) per pixel. In this case the RLE data should be read into an array of shorts. To expand data, the low order seven bits of the first short: bits[6..0] are used to form a count. If bit[7] of the first short is 1, then the count is used to specify how many shorts to copy from the RLE data buffer to the destination. Otherwise, if bit[7] of the first short is 0, then the count is used to specify how many times to repeat the value of the following short, in the destination. This process proceeds until a count of 0 is found. This should decompress exactly XSIZE pixels. Note that the byte order of short data in the input file should be used, as described above. Implementation notesImplementation of both RLE and VERBATIM format for images with BPC of 1 is required since the great majority of SGI images are in this format. Support for images with a 2 BPC is encouraged. If the ZSIZE of an image is 1, it is assumed to represent B/W values. If the ZSIZE is 3, it is assumed to represent RGB data, and if ZSIZE is 4, it is assumed to contain RGB data with alpha. The origin for all SGI images is the lower left hand corner. The first scanline (row 0) is always the bottom row of the image. Naming ConventionsOn SGI systems, SGI image files end with the extension .bw if they are B/W images, they end in .rgb if they contain RGB image data, and end in .rgba if they are RGB images with alpha channel. Sometimes the .sgi extension is used as well. An exampleThis program will write out a valid B/W SGI image file: #include "stdio.h" #define IXSIZE (23) #define IYSIZE (15) putbyte(outf,val) FILE *outf; unsigned char val; { unsigned char buf[1]; buf[0] = val; fwrite(buf,1,1,outf); } putshort(outf,val) FILE *outf; unsigned short val; { unsigned char buf[2]; buf[0] = (val>>8); buf[1] = (val>>0); fwrite(buf,2,1,outf); } static int putlong(outf,val) FILE *outf; unsigned long val; { unsigned char buf[4]; buf[0] = (val>>24); buf[1] = (val>>16); buf[2] = (val>>8); buf[3] = (val>>0); return fwrite(buf,4,1,outf); } main() { FILE *of; char iname[80]; unsigned char outbuf[IXSIZE]; int i, x, y; of = fopen("example.rgb","w"); if(!of) { fprintf(stderr,"sgiimage: can't open output file\n"); exit(1); } putshort(of,474); /* MAGIC */ putbyte(of,0); /* STORAGE is VERBATIM */ putbyte(of,1); /* BPC is 1 */ putshort(of,2); /* DIMENSION is 2 */ putshort(of,IXSIZE); /* XSIZE */ putshort(of,IYSIZE); /* YSIZE */ putshort(of,1); /* ZSIZE */ putlong(of,0); /* PIXMIN is 0 */ putlong(of,255); /* PIXMAX is 255 */ for(i=0; i<4; i++) /* DUMMY 4 bytes */ putbyte(of,0); strcpy(iname,"No Name"); fwrite(iname,80,1,of); /* IMAGENAME */ putlong(of,0); /* COLORMAP is 0 */ for(i=0; i<404; i++) /* DUMMY 404 bytes */ putbyte(of,0); for(y=0; y<IYSIZE; y++) { for(x=0; x<IXSIZE; x++) outbuf[x] = (255*x)/(IXSIZE-1); fwrite(outbuf,IXSIZE,1,of); } fclose(of); } |