初级视皮层机制启发下的特征提取

A Feature Extraction Inspired by Primary Visual Cortex Mechanism

针对传统计算机视觉方法在纹理细节提取和局部运动感知方面的局限性,借鉴生物视觉系统的强大信息处理机制,提出一种基于三维时空能量模型的初级视皮层(V1)视觉模型。模型首先利用三维Gabor滤波器初步模拟得到简单细胞感受野,接着采用半波整流操作进一步得到简单细胞模型,随后通过能量模型整合简单细胞感受野进行,模拟得到复杂细胞模型。仿真结果表明:所提模型在图像静态纹理特征和简单视频序列运动特征提取上取得了较好的效果。较传统模型而言,模型将计算机视觉与生物视觉有效融合,具有更强的动静态特征表达提取能力,体现了V1细胞强大的信息处理机制,为视觉脑启发计算提供新的思路。

In view of the limitation by the traditional methods in texture feature extraction and local motion perception,and referred to the powerful information processing mechanism of human visual system,a primary visual cortex(V1)vision percention model based on 3 Dspatio-temporal motion energy model is proposed.Firstly,simple cell receptive field is simulated by using 3 DGabor filter,then by half wave rectification operation,and further simple cell model is obtained.Finally,the simple cell receptive field is integrated by energy model,so that complex cell model is simulated.The simulation results show that the model possess good performance at extracting static texture feature of images and the local motion feature of simple video sequences.Compared with the traditional model,the model combines computer vision well with bio-vision,having stronger abilities at feature representation and extraction,and showing more powerful information processing mechanism of V1 cells.And this also provides new ideas for brain-inspired visual computing.