Single image super resolution using fuzzy deep convolutional networks

Abstract

Stimulated by the current advancements in Convolutional Neural Networks, a fuzzy deep learning algorithm for Single Image Super Resolution is proposed in this paper. A novel approach is proposed where a fuzzy rule layer is convoluted with deep network to reconstruct a high resolution image. However, the method exploits rule-driven patch selection to directly learn a feature mapping between the input image to super resolved images adopting the advantages of neuro-fuzzy models. The proposed method has been compared with traditional as well as advanced image super resolution techniques. Based on the quantitative and qualitative performance analysis, it is established that our proposed Fuzzy Deep Learning based method is suited for single image super resolution.