Adaptive JPEG Steganography using convolutional codes and synchronization bits in DCT domain

In this paper we propose an adaptive data hiding method that divides the host image in suitable and ineligible blocks. This classification is based on the DCT energy features from the horizontal, vertical and diagonal frequency information. Only the suitable blocks are used for data embedding using quantization index modulation (QIM). After the composite image is attacked by JPEG compression, a desynchronization happen due to positive and negative fail detection in the decoder. For that reason, we propose to use two synchronization bits to solve the desynchronization problem and convolutional codes for decoding errors. Our results show that we can obtain composite images (stego-images) with an average PSNR equal to 36.69 dB and the secret information can be recovered without errors after compression with a quality factor (QF) greater or equal to JPEG quantization matrix quality factor used in the embedding process.