A Model of Saliency-Based Visual Attention for Rapid Scene Analysis
IEEE Transactions on Pattern Analysis and Machine Intelligence
The handbook of brain theory and neural networks
The handbook of brain theory and neural networks
| Hi-index | 0.00 |
In order to improve visual pattern recognition capability, this paper focuses on top-down selective attention at feature space. The baseline recognition system consists of local feature extractors and a multi-layer Perceptron (MLP) classifier. An attention layer is added just in front of the multi-layer Perceptron. Attention gains are adjusted to cope with the top-down attention process and ellucidate expected input features. After attention adaptation, the distance between original input features and expected features becomes an important measure for the confidence of the attended class. The proposed algorithms improves recognition accuracy for handwritten digit recognition tasks, and is capable of recognizing 2 superimposed patterns one by one.