脑皮层视觉认知量化模型及其在目标检测定位中的应用

Quantitative Model of Visual Cortex and its Application in Object Detection

MIT大学的Rieseshuber和Poggio提出了脑颞叶皮层视觉认知系统的标准量化模型(QMVC),这种前馈式层级模型很好地仿效了脑皮层视觉从简单认知元到复杂认知元的识别机理.本文由QMVC模型提取出一组新的特征矢量,这组特征矢量具有对目标变换的不变性.基于QMVC模型的特征矢量建立了新的目标识别系统结构,新目标识别系统对各类目标具有不错的识别率和ROC特性.最后本文引入了尺度窗技术,将新特征应用于复杂场景中的目标检测和定位,实验结果说明本文的新目标检测方法是有效的.

Rieseshuber and Poggio from MIT proposed a standard quantitative model of visual cortex （QMVC）. This feedforward hierarchical model perfect follows the visual recognition mechanisms of the cortex from simple cells to the complex cells. A set of novel features from QMVC which is invariant of object transform can be used in object recognition systems combined with classifiers. This paper proposed a new recognition system framework based on QMVC＇s features. We evaluated the performance of this new system on the single object recognition system and the results show excellent recognition rate and low false alarm rate than other category recognition systems. Together with a scaled window technique, those new features also work well in object detecting and localization in cluster background.