多遮挡场景的光场深度优化估计

Optimal estimation of light field depth for multi-occlusion scenes

光场相机可以同时从多个视点记录场景的方向和光强信息,因此,光场图像具有更大的信息量及更强的可塑性。随着光场技术的发展,针对光场图像的深度估计逐渐成为研究热点,但多遮挡是尚未得到解决的技术难点问题。为此,提出了一种多遮挡场景的光场图像深度优化估计方法。该方法使用优化的高低阈值比的Canny边缘检测算法提取遮挡区域的边缘,在传统的AP聚类算法中引入了迭代加权更新的带加速算子,有效地提高了聚类的精度,较好地解决光场深度估计中的多遮挡问题。采用匹配成本估计初始深度,使用马尔科夫随机场对初始深度结果进行优化处理,得到最终的深度估计结果。实验结果表明,与现有的基于光场图像的深度估计方法相比,本方法不仅抗多遮挡效果显著,而且获得了更低的MSE值,得到了更好的深度估计结果。

The light field camera can record both direction and light intensity information of the scene from multiple viewpoints at the same time, so the light field image has larger amount information and stronger plasticity.The depth optimal estimation of light field image has gradually become a research hotspot with the development of light field technology, but multi-occlusion is still a technical difficulty that has not been solved yet.Therefore, an improved depth estimation method of light field image for multi-occlusion scenes is proposed.In this method, Canny edge detection algorithm with optimized high-low threshold ratio is used to extract the edge of occluded area.An iterative weighted update operator with acceleration is introduced into the traditional AP clustering algorithm, which effectively improves the accuracy of clustering and solves the multi-occlusion problem for depth estimation of light field.A matching cost is used to estimate the initial depth, and the initial depth results are optimized using Markov random field to obtain the final depth estimation results.The experimental results show that the proposed method not only resists multiple occlusions significantly, but also obtains lower MSE values and better depth estimation results compared with the existing depth estimation method based on light field image.