Decompressing Animated GIF Raster Data

Question

Sorry for the lengthy post I'm trying to give as much information as I can and I did my best to format everything to be as easily readable as possible. I've been trying to decompress GIF's in C# and seem to have everything except the LZW decompression down. I am reading in the Gif from a URL. For this example, I will be using this Animated GIF. Which I know has 35 Frames but I only want to look at the first one.

HEADER: 474946383961
GIF Version: 89a

LOGICAL DESCRIPTOR: 41003D00F60000
Width : 65
Height: 61
Sorted Color Table: False
Global Color Table Size: 128
Background Color Index: 0
Pixel Bits: 0

Below is the Global Color Table, something I am slightly confused about because it is filled 123/128 colors and the rest are 000000s, do I discredit this while filling the code table?

Sorry about the format here, only showing it for the question regarding the 000000s

Global Color Table:
141414 181818 1C1C1C 242424 262626 282828 2A2A2A 2C2C28 302C2C 30302C 3430303 434303
434344 030185 030284 03C346 038306 438306 43C305 044245 048245 448285 C50285C 502C5C
542C5C 543054 503860 542C68 50286C 50286C 542870 542870 542C70 582874 582870 582C745
82C745 C2C785 C2C605 830645 830645 C30645 C34685 C307C6 02C6C6 0346C6 438746 434786
030786 838786 C3C006 464806 02C806 42C806 430846 C34886 C348C7 034847 43C887 0388C7 
43C887 83C8C7 83C907 034947 034947 434987 434947 438987 4389C7 8389C7 C38A07 838A07 
C388C7 C40A46 050A48 03C948 048A08 848A08 C48A89 04CA89 44CAC9 44CAC9 850B09 C50B49 
C50B49 C54B4A 054BCA 050B8A 054B8A 454BCA 454BCA 458BCA C5CC0A 454C0A 854C0A 858C4A 
858C4A C58C4A C5CC8A C58C8A C5CC4B 058C8B 058C8B 05CC89 C64C8B 060CCB 460D0B 860D0B 
864D4B C64D4C 068D8C 068D8C 468DCC 468DCC 86CE8D 070ECD 874F0D C78F4D C78F4D C7CF4E 
078F8E 07CF8E 47C000 000000 000000 000000 000000 000000

Graphics Control Extension

21F9040D03007B00 |Graphics Control Extension
Block Size: 4
Has Transparency: True
Delay: 768
Transparency Color Index: 123

Image Descriptor

IMAGE DESCRIPTOR: 2C0000000041003D0000
Left: 0
Top: 0
Width : 65
Height: 61
Local Color Table: False
Interlace: False

Finally Where I am confused the Image Data

LZW minimum code size: 7
IMAGE DATA SUBBLOCK 1 HEXDATA: 
80 7B 82 83 84 85 86 87 88 89 8A 8B 8C 8D 8E 8F 86 06 06 90 94 84 07 
95 86 05 0C 00 98 94 05 01 07 05 9D 7B 05 0B 9B A3 8F 07 0C 0C 02 05 
93 94 07 08 0A 09 9C A8 8D 05 08 B2 04 02 AF A9 09 0B 0A B5 B6 8B 05 
00 C0 AB 01 A2 8E 05 BF C0 C2 C3 8A 07 0A 08 AB 0F 19 CA 8C A5 C0 0B 
CF D0 88 C5 0C 09 0C 08 6E 29 CB 0C 0B A6 DD DE 87 9A E9 4C 67 2A D8 
89 06 0C 0A E9 08 EB EC 85 06 00 BF 58 63 C6 C8 5B 64 40 4B BA 05 B4 
F6 11 03 F0 20 60 18 37 03 E9 85 41 90 4E 43 05 85 C4 AA 9C 39 43 E6 
8C 0C 15 8A 0A 98 E9 58 A6 0A 05 8C 21 29 90 21 33 86 0C 98 88 99 C0 
8C D9 38 E6 22 CA 90 13 58 86 21 23 05 45 81 9F 40 81 5A 09 C8 F2 E4 
CD 94 1D 37 5C D0 93 A7 A9 D3 A6 2A 7C 8C 34 63 F2 28 B1 09 14 EE D8 
A9 B3 B5 8E 57 AE 76 B8 AC 68 49 66 82 D5 6C 75 E2 A8 55 0B 67 6D 1C 

For All intensive purposes We should need to look at the first few binary bits

IMAGE BLOCK BINARY: 
10000000 clearcode, 01111011, 10000010, 10000011, 10000100, 
10000101, 10000110, 10000111, 10001000, 10001001, 10001010, 
10001011, 10001100, 10001101, 10001110, 10001111,

Codes:
1000000, 0011110, 1110000, 01010000... ect 

My main question is how do I use LSB Packing order when reading these codes, secondly how does this make sense for each pixel considering the background is transparent, like how do I get the index of the first non transparent pixel. Finally, at what point do I increase the code size for adding codes to the table to LZW Minimum codes size +1(8). Thank you for any advice.

Answer 1

LSB packing order just means to read the data as little-endian and right shift the data as you "eat" the bits.

Here's an example in C, C# makes accessing memory more painful, but the logic would be the same:

uint32_t ulBits;
unsigned char *pData;
int codelen, code, bitnum;
int mask;
int nextcode;

codelen = 7; // assume 7 bits to start
mask = (1<<(codelen+1)) -1;
clearcode = (mask >> 1) + 1;
nextcode = clearcode + 2;

ulBits = *(uint32_t)pData; // read 32-bits as little endian
bitnum = 0;
#define WORDLEN 32
// To read the variable length codes you would do the following:

while (decoding == true)
{
   if ((bitnum + codelen) > WORDLEN) // need to read more data
   {
      pData += (bitnum >> 3); // adjust source pointer
      ulBits = *(uint32_t)pData; // read another 32-bits
      bitnum &= 7; // reset bit offset
   }
   code = (ulBits >> bitnum);
   code &= mask;
   bitnum += codelen;
   // some logic here to increment the nextcode is beyond the scope of this answer
   <the rest of your logic here>
}

As you decompress the codes, you add a new item to your dictionary and increment your "next code" value. When this value can't fit in the current code size, you increase it by one bit until you hit 4096 and usually start over with a clear code to reset the dictionary. There is a rarely used option called "differed clear code". In this case the full dictionary stays in use until a clear code is received. There are plenty of sample LZW decoders that you can look at, so it's not necessary to post an entire one here.