Lightweight Feature Fusion Network for Single Image Super-Resolution

Abstract

Single image super-resolution(SISR) has witnessed great progress as\nconvolutional neural network(CNN) gets deeper and wider. However, enormous\nparameters hinder its application to real world problems. In this letter, We\npropose a lightweight feature fusion network (LFFN) that can fully explore\nmulti-scale contextual information and greatly reduce network parameters while\nmaximizing SISR results. LFFN is built on spindle blocks and a softmax feature\nfusion module (SFFM). Specifically, a spindle block is composed of a dimension\nextension unit, a feature exploration unit and a feature refinement unit. The\ndimension extension layer expands low dimension to high dimension and\nimplicitly learns the feature maps which is suitable for the next unit. The\nfeature exploration unit performs linear and nonlinear feature exploration\naimed at different feature maps. The feature refinement layer is used to fuse\nand refine features. SFFM fuses the features from different modules in a\nself-adaptive learning manner with softmax function, making full use of\nhierarchical information with a small amount of parameter cost. Both\nqualitative and quantitative experiments on benchmark datasets show that LFFN\nachieves favorable performance against state-of-the-art methods with similar\nparameters.\n