Two-stream encoder-decoder network for localizing image forgeries

This paper proposes a novel two-stream encoder–decoder network that utilizes both the high-level and low-level image features for precisely localizing forged regions in manipulated image. is motivated by fact forgery creation process generally introduces artefacts ( e.g. , unnatural contrast) noise inconsistency) to images. In proposed network, one stream learns manipulation-related encoder side extracting residuals through set of high-pass filters first layer. second stream, manipulation from input RGB values. The coarse feature maps each are upsampled corresponding decoder produce dense maps. two streams concatenated fed final convolutional layer with sigmoidal activation pixel-wise prediction. We have carried out experimental analyses on multiple standard forensics datasets evaluate performance method. results show efficacy method respect state-of-the-art.