太赫兹光场成像的深度估计方法与实验研究

在前期工作中,通过对太赫兹光场图像进行离散余弦变换(Discrete Cosine Transform,DCT)滤波和数字重聚焦,初步实现了图像去噪和前后景分割。为了进一步得到质量更高的太赫兹光场原数据并达到更加精确的深度分割效果,改进了实验方案及处理方法,并提出了一种基于极平面图像(Epipolar Plane Image,EPI)的太赫兹光场深度估计方法。在太赫兹图像特性的基础上,给出了深度与视差的关系,并利用局部视差和置信度构建了全局深度图,从而达到了深度估计的目的。最后,在实验中通过10×10的相机阵列采集太赫兹光场数据,得到了准确聚焦于不同平面的重聚焦结果和高分辨度的深度估计图,实现了太赫兹光场成像的深度估计。

Research on Depth Estimation Method and Experiment of Terahertz Light Field Imaging

In the previous work,through discrete cosine transform and digital refocusing,the terahertz optical field image preliminarily realized image denoising,as well as foreground and background segmentation.In order to further obtain higher quality terahertz optical field original data and achieve more accurate depth segmentation,the experimental scheme and processing method are improved,and a terahertz optical field depth estimation method based on epipolar plane image is proposed in this paper.On the basis of terahertz image characteristics,the relationship between depth and disparity is given,and the global depth map is constructed by using local parallax and confidence,so as to achieve depth estimation.Finally,experiments are performed by collecting terahertz light field data from camera array.Results of accurate refocusing on different planes and a high-resolution depth estimation map are obtained,and the depth estimation of terahertz optical field imaging is realized.