Apparatus and method for executing deep learning based watermark in various kinds of content environment

摘要

The apparatus for executing a deep learning-based watermark in a multi-content environment of the present invention includes an artificial neural network including a plurality of layers input to the calculation of the next layer by applying a weight to the calculation result of the previous layer, and the original content. , Protected content in which a watermark is mapped to the original content, and a plurality of damaged protected content in which a plurality of different watermark attacks are performed by executing a plurality of different watermark attacks to damage the watermark of the protected content. A learning unit for deriving a content weight, which is a weight of the artificial neural network, to determine that the original content, the protected content, and the plurality of damaged protected content are the same by learning the artificial neural network using the learning data set; And a mapping unit configured to generate a weighted watermark by calculating an exclusive logical sum of a content weight and a watermark, and then mapping the weighted watermark to an original of the content to generate a protected content.