仿视皮层机制的随机点视频序列运动特征提取

Motion Feature Extraction of Random-Dot Video Sequences with the Visual Cortex Mechanism

  • 摘要: 为了探索视皮层对多方向刺激的运动信息处理机制,依据生物实验中部分中颞叶区(MT)细胞对随机点视频刺激运动感知的侧偏特性,提出了一种仿视皮层机制的随机点视频序列运动特征提取模型。首先,采用Von Mises函数拟合初级视皮层(V1)简单细胞的感受野,再结合非线性的能量模型、调谐和非调谐正则化方法等模拟实现V1复杂细胞对运动信息的方向感知;其次,提出了级联前馈方法对V1复杂细胞响应进行线性加权求和,得到MT细胞的响应输出;最后,利用生物实测MT细胞响应数据对新模型的有效性进行了实验验证。仿真结果表明模型的输出结果与恒河猴的运动感知实验数据基本吻合,能够较好地模拟视皮层对不同夹角的多方向随机点视频序列的运动感知机制,为提取复杂运动序列的特征提供新的方法,进而为视觉类脑计算奠定基础。

     

    Abstract: According to side-bias characteristics of motion perception towards random-dot video stimuli among some middle temporal cortex (MT) cells in the experiment, a model for motion feature extraction of random-dot video sequences with the visual cortex mechanism is proposed for exploring the visual cortex mechanism underlying multi-direction motion information processing. First, Von Mises function is used to model receptive fields of simple cells in primary visual cortex (V1). A nonlinear energy model unturned and tuned normalization operation is adopted to simulate the direction perception of V1 complex cells; then, with the feedforward and cascade method, the responses of V1 model are pooled through a set of linear weights, thus giving rise to the responses of MT cells. The simulation results are generally consistent with the experimental data of rhesus monkeys. It can model the motion perception and feature extraction mechanism underlying multi-direction random-dot video sequences at different direction separations in visual cortex. So that it can provide new ideas for extracting features of complex motion sequences and lay the foundation for the visual brain-like computation for the further step.

     

/

返回文章
返回