Camouflaged Object Segmentation Based on Fractional Edge Perception

Camouflaged object detection is a challenging task because of intrinsic similarity between background and foreground. In the existing models, many confusing edge features eventually lead to bad predictions. In this paper, we propose an Interactive Task Learning Network, in which an improved fractional-order differential operator is used to calculate the gradient intensity of the image. By calculating average gradient and spatial frequency of image, it can adaptively learn the fractional-order v and extract features from different directions. Experiments on three camouflaged datasets indicate that the proposed method is effective and progressiveness.