Damage-less robust watermark extraction using non-linear feature extraction scheme trained on frequency domain

In this paper we propose a new information hiding and extracting method without embedding any information to a target content using non-linear feature extraction trained on frequency domain. Our system can detect a hidden bit code from the content by processing coefficients of the selected feature block of frequency domain. For the generation of the keys which is needed for extracting a bit code from a content, is done by a supervised learning of the set of values in the selected feature block with the teacher signal value. The teacher signal value is the bit code that you want to relate to the content. The connection weight which was processed by the supervised learning will be used as the key for extracting the bit code that you related to the content. With our proposed method, we were able to introduce a watermark scheme with no damage to a target content because there are no information added to the target content, and this characteristic is effective when you don’t want the target content to be damaged at all. Key-Words: damage-less embedding, non-linear feature extraction, supervised learning on DCT domain, robust watermark, feature selection on frequency domain