Meow Transmission Revenge - Forensics

Points: 451 | Flag: BITSCTF{r3v3ng3_0f_th3_c4t_m4p_m4st3r} | Solved by: Smothy @ 0xN1umb

what we got

a single file revenge_transmission.png - 128x128 RGB image. challenge says its "chaotic noise" and the cat has a "triple personality" now. metadata is packed with Arnold Cat Map parameters.

the solve

step 1: check the metadata

exiftool revenge_transmission.png

metadata was absolutely LOADED:

ok so Arnold Cat Map with different params per RGB channel. 3 stages each ("leaps"), different matrix styles and iteration counts ("spins"), applied in a specific order ("sequence"). got it.

step 2: the big bamboozle

spent way too long trying to "unscramble" the full image thinking it was noise. turns out... it wasn't noise at all lmao. the image is literally just a picture of a cat. mean horizontal pixel difference was only 2.9 - that's smooth af.

the REAL data was hiding in the LSBs (least significant bits) of each channel. classic stego move ngl.

# extract LSBs
for ch_idx, ch_name in enumerate(['R', 'G', 'B']):
    lsb = (pixels[:, :, ch_idx]) & 1
    print(f"{ch_name} LSB: {np.sum(lsb)} ones out of {N*N}")

exactly 319 white pixels per channel. very sparse, very intentional. the LSB planes showed scattered dots - a scrambled binary image.

step 3: arnold cat map parameters

had to figure out what "Styles" meant. after trying like 5 different interpretations:

• Style 1: standard Arnold Cat Map matrix [[1,1],[1,2]] → (p=1, q=1)
• Style 2: generalized Arnold Cat Map [[1,2],[1,3]] → (p=2, q=1)

the Arnold Cat Map formula being:

Sequence determines the order stages are applied. e.g. Red's [1,2,3] = apply stage 1 first, then 2, then 3. Green's [2,3,1] = stage 2 first, then 3, then 1.

step 4: unscramble the LSBs

for each channel:

1. extract the LSB plane
2. build the forward stage order from Sequence
3. apply the inverse Arnold Cat Map for each stage, in reverse order

style_map = {
    1: (1, 1, 1, 2),  # [[1,1],[1,2]]
    2: (1, 2, 1, 3),  # [[1,2],[1,3]]
}

for ch_idx, ch_name in enumerate(['R', 'G', 'B']):
    cfg = configs[ch_name]
    lsb = (pixels[:, :, ch_idx] & 1).astype(np.uint8)

    # build forward order from sequence
    stages = []
    for seq_val in cfg['sequence']:
        idx = seq_val - 1
        style = cfg['styles'][idx]
        spins = cfg['spins'][idx]
        stages.append((style_map[style], spins))

    # apply inverse in reverse order
    decoded = lsb.copy()
    for (a, b, c, d), spins in reversed(stages):
        # inverse of [[a,b],[c,d]] with det=1 is [[d,-b],[-c,a]]
        inv = (d % N, (-b) % N, (-c) % N, a % N)
        for _ in range(spins):
            decoded = cat_scatter(decoded, *inv, N)

step 5: profit

the unscrambled LSBs revealed pixel-art text across all three channels:

key takeaways

• dont assume the image is scrambled just bc the challenge says so. CHECK THE PIXELS FIRST
• arnold cat map on the LSBs, not the full image - sneaky af
• "triple personality" = different cat map params per RGB channel
• "Styles" = which (p,q) parameter set. Style 1 = (1,1), Style 2 = (2,1)
• "Spins" = iteration count per stage
• "Sequence" = order to apply the stages (1-indexed)
• to reverse: apply inverse matrix for each stage in reverse sequence order

flag

BITSCTF{r3v3ng3_0f_th3_c4t_m4p_m4st3r}

revenge of the cat map master - respect tbh

smothy out ✌️