基于视觉感受野的自适应背景抑制方法  

Adaptive background suppression method based on visual receptive field

作　　者：张振[1] 陈哲[1] 吕莉[1] 王鑫[1] 徐立中[1] 

机构地区：[1]河海大学计算机与信息学院,南京211100

出　　处：《仪器仪表学报》2014年第1期191-199,共9页Chinese Journal of Scientific Instrument

基　　金：国家自然科学基金(61263029;61374019);江苏省自然科学基金(BK20130851)资助项目

摘　　要：在河流水面成像测速的应用中,表现为弱小目标的水流示踪物易受倒影、耀光等复杂背景噪声的干扰而引起较大的位移估计误差。对此,首先分析了近红外河流水面图像中目标、背景及噪声的分布及统计特性并建立数学模型。然后在生物侧抑制现象的启发下提出了一种基于视觉感受野双高斯差(DOG)模型的自适应背景抑制方法。利用水面图像中目标和噪声灰度分布的先验知识以及兴奋性与抑制性作用相抵的约束关系选取模型参数,以达到局部最优的增强效果。实验表明,DOG模型作为一个带通滤波器,在增强目标、抑制背景和滤除噪声的综合性能方面优于传统的空域高通滤波器。获得的图像不仅具有良好的视觉效果,同时满足了后续运动矢量估计对相关运算信噪比的需求。In the application of fiver surface imaging velocimetry, the water flow tracers shown as dim and small targets are easily affected by complex background noises, such as shadows and reflections, which leads to large errors in dis- placement estimation. To solve this problem, firstly the distribution and statistics characteristics of target, background and noise in NIR fiver surface images are analyzed to build the mathematical model. Then, inspired by the biological phenomenon of lateral inhibition, an adaptive background suppression method is presented based on the visual receptive field difference of Gaussian （DOG） model. To achieve local optimization of enhancement,the model parameters are de- termined using the prior knowledge of the intensity distributions of targets and noises in the fiver surface images, and the constraint relation that the excitatory and inhibitory effects compensate for each other;and the local optimization en- hancement effect is achieved. The experiment results show that, as a band-pass filter, the DOG model is superior to tra- ditional spatial high-pass filter in the performance of target enhancement, background suppression and noise fihering. The images obtained with the proposed method not only have good visual effects, but also meet the requirement of suffi- cient signal-noise ratio （SNR） for the correlation operations in subsequent motion vector estimation.

关 键 词：河流水面 光学环境 近红外成像 背景抑制 视觉感受野 双高斯差 

分 类 号：TP751[自动化与计算机技术—检测技术与自动化装置]