| Abstract: | The current paper describes a new technique for embedding secret data in JPEG compressed color digital images. Emphasis is given in improving effective payload percentage over file size increase while retaining low distortion levels. The proposed technique is based on modification of non-zero quantized coefficients. In contrast to previously proposed techniques, the present one does not require the use of non-standard Huffman tables. Apart from that, no side information needs to be included in a JPEG header to assist the decoder in extracting the hidden message. The main idea behind the technique is to create zero population bins in the coefficient histogram in order to accommodate the secret message bits. Specifically, all coefficients whose absolute value is L are modified by 1 in the direction of their sign when a bit 1 is to be embedded or are not modified when a bit 0 is to be embedded. Coefficients with absolute values greater than L are also incremented by one in the direction of their sign, so that they are not confused with coefficients where embedding has occurred. At the side of the decoder, both the message bits can be extracted from coefficients with absolute values L and L+1, and the original image can be restored by shifting all coefficients with absolute value greater than L in the opposite direction of their sign. The results show significant improvement in comparison with representative state-of-the-art techniques, in terms of ratio of payload over file size increase. |
