基于层次特征映射模型的目标识别

Object recognition based on a hierarchical feature map model

为了较好地模拟生物视觉系统对复杂场景目标的感知特性,以提高目标识别水平,提出了一种新的受生物视觉信息处理的基本机理启发的前馈深度层次计算机视觉模型,即层次特征映射(HFM)模型。该模型利用高斯差分函数以及Gabor函数模拟初级视觉皮层中的方向图,并且采用竞争学习策略来学习更高层次神经元的感受野。实验表明,该模型可以很好地提取目标的特征和保留图像的主要信息,并且具有自学习的能力,能够在主流数据库上取得较好的识别结果,具有较好的发展前景。

A new biologically inspired feed-forward deep hierarchical model,i.e.the hierarchical feature map （HFM） model,is introduced to better simulate the biological vision system＇ s perception of objects in a complex scene for improvement of the visual object recognition.The HFM model uses the Difference of Gaussian function and Gabor function to simulate the orientation map in the primary visual cortex V1,and adopts a competitive learning strategy to learn the receptive field （RF） properties of higher level neurons.The experimental results show that the HFM model could well preserve the main structure of images.The model is also capable of self-taught learning and can achieve promising results on popular image databases,showing a good prospect for development.