When Dictionary Learning Meets Deep Learning: Deep Dictionary Learning and Coding Network for Image Recognition With Limited Data

We present a new deep dictionary learning and coding network (DDLCN) for image-recognition tasks with limited data. The proposed DDLCN has most of the standard layers (e.g., input/output, pooling, fully connected), but fundamental convolutional are replaced by our compound layers. learns an overcomplete input training At layer, locality constraint is added to guarantee that activated bases close each other. Then, atoms assembled passed In this way, in first layer can be represented deeper second dictionary. Intuitively, designed learn fine-grained components shared among atoms; thus, more informative discriminative low-level representation obtained. empirically compare several leading methods models. Experimental results on five popular data sets show achieves competitive compared state-of-the-art when limited. Code available at https://github.com/Ha0Tang/DDLCN.