Steganography is the art and science of hiding secret information in a cover medium such that the presence of the hidden information cannot be detected. This thesis proposes a new method of steganography by cover modification in JPEG images. Essentially, the algorithm exercises LSB replacement using the definition for steganographic values from F5. After the nonzero quantized DCT coefficients of a cover image undergo a pseudorandom walk, the coefficients and the payload are split into an equal number of partitions and paired. Each coefficient partition is permuted again by the 1/P pseudo-random number generator until an optimal embedding efficiency for its corresponding payload is achieved. Using this method, we achieve a higher embedding efficiency than that of LSB replacement alone. We evaluate the detectability of our algorithm by creating a multi-classifier based on the output of multiple non-linear, soft-margin support vector machines trained on POMM features. We show that our algorithm performs nearly as well as the state-of-the-art nsF5 algorithm, and outperforms other state-of-the-art algorithms under most conditions.