视觉注意驱动的基于混沌分析的运动检测方法

提出了视觉注意驱动的基于混沌分析的运动检测方法(MDSA)。MDSA首先基于视觉注意机制提取图像的显著区域，而后对显著区域进行混沌分析以检测运动目标。算法技术路线为:首先根据场景图像提取多种视觉敏感的底层图像特征；然后根据特征综合理论将这些特征融合起来得到一幅反映场景图像中各个位置视觉显著性的显著图；而后对显著性水平最高的图像位置所在的显著区域运用混沌分析的方法进行运动检测；根据邻近优先和返回抑制原则提取下一最显著区域并进行运动检测，直至遍历所有的显著区域。本文对传统的显著区域提取方法进行了改进以减少计算量:以邻域标准差代替center-surround算子评估图像各位置的局部显著度，采用显著点聚类的方法代替尺度显著性准则提取显著区域；混沌分析首先判断各显著区域的联合直方图（JH）是否呈现混沌特征，而后依据分维数以一固定阈值对存在混沌的JH中各散点进行分类，最后将分类结果对应到显著区域从而实现运动分割。MDSA具有较好的运动分割效果和抗噪性能，对比实验和算法开销分析证明MDSA优于基于马塞克的运动检测方法（MDM）。 

Motion detection driven by visual attention mechanism applying chaos analysis method

A motion detection technology driven by visual attention mechanism applying chaos analysis method is proposed in this paper. The method firstly extracts salient regions of the scene image based on visual attention mechanism, and then detects motion objects with chaos analysis method on salient regions. The technical route of the motion detection technology driven by visual attention mechanism applying chaos analysis method is as follows: firstly, bottom image features that are sensitive to vision are extracted from scene image; secondly, salient map which reflects the visual saliency of each scene location is obtained via incorporating these image features according to the feature integration theory; and then, motion detection is done with the chaos analysis method on the salient image region which contains the most salient scene image location; finally, the next salient image region with the strongest saliency in the residual ones is detected according to the proximity criterion and the inhibition-of-return criterion; the process given above is repeated till having detected motion objects on all scene image regions. To decrease computational complexity, the traditional method of extracting salient image region is improved as follows: local standard deviation operator replaces center-surround operator to estimate local saliency of each image location and salient pixel clustering method replaces scale saliency rule to extract salient regions in our method. The chaos analysis method firstly estimates whether the joint histogram puts up chaotic characteristics, then classifies all scatters of the joint histogram that puts up chaotic characteristics by fractal dimension with a fixed threshold, and finally corresponds the classified result to salient regions to segment motion objects. Motion detection technology driven by visual attention mechanism is effective and robust. The contrast experiments and algorithm cost analysis are done which show that our method excels the motion detection method based on mosaics in segmentation effect and velocity.