基于脉冲余弦变换的选择性视觉注意模型

Selective Visual Attention Model Based on Pulsed Cosine Transform

提出一种基于脉冲余弦变换的视觉注意模型,它模仿自底向上视觉注意的形成机制.该模型结构简单,计算速度快,能够应用于实时处理系统.在该模型中,视觉显著性可表示为二元编码,这与人脑神经元脉冲放电方式相符合.运动显著性也可通过这些二元编码生成.此外,该模型还可推广为基于Hebb学习规则的神经网络.实验结果表明,在人眼注视点预测性能上,该模型优于其它经典视觉注意模型.

A visual attention model based on pulsed cosine transform is proposed, which mimics the generating mechanism of bottom-up visual attention. Due to its simple architecture and high computational speed, the proposed model can be used in real-time systems. The visual salience of the model is represented in binary codes, which agrees with the firing pattern of neurons in the human brain. The motion salience is generated by these binary codes as well. Moreover, the model can be extended to Hebbian-based neural networks. Experimental results show that the proposed model has better performance in human fixation prediction than other state-of-the-art models of visual attention.