基于生物视觉通路的目标识别算法

研究哺乳动物视觉通路的结构和功能,为机器学习提供了广泛的思路。文章对经典稀疏编码和HMAX模型进行改进,建立一种模拟完整视觉通路的算法。用4DGabor金字塔模拟了视觉信息从视网膜到腹侧通路V1区的处理过程;设计一种带稀疏编码性质的非线性滤波器,模拟了信息在V1区到PFC区的多层次处理步骤。实验表明该算法能够符合已知生物模型,达到现有同类先进算法的效果。

Object recognition algorithm based on biological visual pathway

The research on the anatomical and functional connectivity of visual pathway affords a broad way of machine learning.An algorithm to simulate the whole visual pathway is presented based on the improved classic SC and HMAX models.This process can be broken down into two steps: the first is a coding step,which utilizes 4D Gabor pyramid to simulate visual information processing from the retina to the ventral pathway V1 area,and the second is a pooling step,which utilizes a sparse nonlinear filter to simulate multi-level visual information processing from V1 area to PFC area.The experimental results show that this approach tallies with the living model and achieves the result of the state-of-the-art model.