Analysis Dictionary Learning: An Efficient and Discriminative Solution

This repository contains the demo and MATLAB codes for our conference paper: "Analysis Dictionary Learning: An Efficient and Discriminative Solution" by Wen Tang, Ashkan Panahi, Hamid Krim and Liyi Dai, which is published in ICASSP 2019.

Citation

If you think our project is useful, please consider citing it:

@inproceedings{tang2019analysis,
  title={Analysis Dictionary Learning: an Efficient and Discriminative Solution},
  author={Tang, Wen and Panahi, Ashkan and Krim, Hamid and Dai, Liyi},
  booktitle={ICASSP 2019-2019 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP)},
  pages={3682--3686},
  year={2019},
  organization={IEEE}
}

Introduction

Discriminative Convolutional Analysis Dictionary Learning (DCADL) method, as a lower cost Discriminative DL framework, to both characterize the image structures and refine the interclass structure representations. The proposed DCADL jointly learns a convolutional analysis dictionary and a universal classifier, while greatly reducing the time complexity in both training and testing phases, and achieving a competitive accuracy, thus demonstrating great performance in many experiments with standard databasets.

Discriminative Convolutional Analysis Dictionary Learning

An intuitive DCADL frameworkis defined as follows:

Noting that conventional ADL formulations rely on matrix multiplication for efficient solution, we reformulate above convolutional ADL problem by assuming that images have no zero-padding. In this case, we segment an image into patches with pixels, being of the same size as the atom, and let and .

The above DCADL can then be rewritten in the same form as the one of the conventional ADL with two reshaping operators with :

Such kind of transformation greatly reduce the excessive training time and to ensure the capability of characterizing structuresamong individual images and across classes.

Contacts

email: wtang6@ncsu.edu