Ada-Sal Network: emulate the Human Visual System

作者：

Highlights：

• We propose Sal-Mask Connection to mimic visual attention mechanism.

• We deduced formulas for feedforward and backpropagation of the Sal-Mask Connection.

• We visualized the feature maps to illustrate the efficiency of Sal-Mask Connection.

• We proposed Ada-Sal Network to learn better saliency and get better performances.

摘要

Highlights•We propose Sal-Mask Connection to mimic visual attention mechanism.•We deduced formulas for feedforward and backpropagation of the Sal-Mask Connection.•We visualized the feature maps to illustrate the efficiency of Sal-Mask Connection.•We proposed Ada-Sal Network to learn better saliency and get better performances.

论文关键词：Ada-Sal Network,Convolutional Neural Network,Human Visual System,Image classification,Machine learning,Saliency,Sal-Mask Connection

论文评审过程：Received 15 October 2015, Revised 27 May 2016, Accepted 28 May 2016, Available online 24 June 2016, Version of Record 29 September 2016.

论文官网地址：https://doi.org/10.1016/j.image.2016.05.017