Unable to decompress Xref Stream

Hi

I am trying to get the cross-reference stream for weeks now, and have almost pulled all my hair out. =(
Can anyone show me how to get the cross-reference stream?
I use the FlateDecode() from iText first, then i applied the filter algorithm.

But the results (in hex) i got are weird:

stream after flatedecode
-----------------------------
02 010000 10
00 020000 02
8A 000200 00
01 9A0002 00
00 01EB00 02
00 00014E 00
02 000001 15
00 020001 7D
69 000200 00
05 170002 00
00 01C500 02
00 00013E 00
02 00FF76 6F
00 020100 00
AB 000200 00
00 000102 00
00 000001 02
00 000000 01
02 000000 00
01 020000 00
00 010200 00
00 000102 00
00 000001 02
00 000000 01
02 000000 00
01 020000 00
00 010200 00
00 000102 FF
00 01BAF5

stream after filtering
------------------------
02 010000 10
FE 010000 F2
8C FF0200 0E
75 9BFE02 F2
8B 66EDFE 10
75 9A1450 F0
8D 66ECB1 25
73 9C1450 58
F6 64EEB0 A8
0F B31252 58
F1 4EB3AE AA
0F B24E90 56
F3 4EB1E6 19
0D B2501A E7
9E 4EB2E6 19
62 B24F1C E7
9E 4EB1E5 1B
62 B24F1B E6
A0 4EB1E5 1A
61 B44F1B E6
9F 4DB3E5 1A
61 B34E1D E6
9F 4DB2E4 1C
61 B34E1C E5
A1 4DB2E4 1B
60 B54E1C E5
A0 4CB4E4 1B
60 B44D1E E4
A0 4D6DD7

The decodeParms are "<</DecodeParms<</Columns 5/Predictor 12>>"
Therefore to reverse filter, for every byte, i minus the byte above it. (i even tried adding). But the results does not seem correct.
I am expecting that the 1st column should be either 0,1 or 2 (according to pdf specification).

Can anyone please help???
Thanks in advanced!