I have been looking for image encryption after compression using Ax+e, where A is a random matrix, x is plaintext and b is a bitstream generated using a stream cipher. Suppose choosing plaintext as all zeroes (dark image), will reveal many keystream bits. With the sufficiently large amount of known keystream bits, it is possible to obtain the original key. Is there any upper bound on the number of known keystream bits for a stream ciphers that can guarantee a sufficient resistance against attacks? For example, if the sequence has been generated by an LFSR of length n, then the Berlekamp-Massey algorithm is guaranteed to find this LFSR after examining no more than 2n bits of the sequence.
Is it possible to make above system cryptographically secure against chosen plaintext attack?
Is changing the initiation vector frequently so that attacker can get the limited amount of keystream bits only solution?
How to use initialization vector in such scenario?
Is one-time permutation after obtaining keystream bits proper solution?