CAJ | 학술논문

为了寻求代价更小、效率更高、适应性更强的图像局部特征表征方法，提出一种基于视觉机制的多层网络计算模型。首先对初级视皮层中的简单细胞和复杂细胞等神经元进行建模；然后对腹侧视通路上的V4区神经元和下颞叶皮层区神经元的响应模式进行研究，并利用该计算模型对输入图像进行局部特征的表征。实验结果表明，与传统的图像特征描述方法相比，该模型所提取的图像局部特征具有足够的区分度；此外，利用生物视觉模型提取出的图像局部特征在具有复杂背景的场景中显示出了更加优秀的泛化能力。
위료심구대개경소、효솔경고、괄응성경강적도상국부특정표정방법，제출일충기우시각궤제적다층망락계산모형。수선대초급시피층중적간단세포화복잡세포등신경원진행건모；연후대복측시통로상적V4구신경원화하섭협피층구신경원적향응모식진행연구，병이용해계산모형대수입도상진행국부특정적표정。실험결과표명，여전통적도상특정묘술방법상비，해모형소제취적도상국부특정구유족구적구분도；차외，이용생물시각모형제취출적도상국부특정재구유복잡배경적장경중현시출료경가우수적범화능력。
For representing local image features, minor price, more efficient and more flexible, a hierarchical network model based on human vision physiological mechanism was put forward. Firstly, simple cell and com-plex cell in primary visual cortex are modeled, then studied the response pattern of V4 area and inferior temporal cortex on ventral side channel and representing the local features of input image utilized the computational model. The experiment results show that local image features extracted by computational model have sufficient dis-crimination; furthermore, the local image features extracted using biological visual model demonstrated much more excellent generalization ability in natural scene with complicated background.