核相关滤波在散斑图像相关分析的应用研究

Research on the Application of Speckle Image Correlation Analysis Based on Kernel Correlation Filter

进行参考图像与目标图像的匹配是数字散斑图像相关分析中很关键的一个环节,完成这一任务的主流方法是使用传统的互相关工具来评估局部图像原始像素数据之间的相关度,以得到的结果来确定匹配的位置。鉴于相关滤波器在计算机视觉的目标跟踪领域有良好的性能表现,而核函数又可以作为获取图像数据更深层次特征的桥梁,核相关滤波算法被作为一种新的相关度评估工具被应用于数字散斑图像的相关分析。在一定椒盐噪声和尺寸压缩的条件下,实验数据表明,这一方案具有准确性高和性能稳定的特点。

For digital speckle image correlation analysis,it is one of the critical steps to find out the match point between the reference image and tar⁃get image.The common approach for this task is that making use of the traditional cross-correlation as a tool to estimate the correlation of raw data between sub-windows from reference image and target image.And the match point is located depending on the correlation value.Correlation filter has been proved as an effective solution for object tracking in computer vision,and kernel trick bridges the gap between raw data and its high-dimensional features.Kernel correlation filter may be a better method for speckle image correlation analysis.Experi⁃mental result shows that kernel correlation filter has advantage of accuracy and robust under a range of impulse noise and scaling which is close to realistic situation.