Abstract:

To solve the low accuracy problem caused by occlusion interference of light field depth estimation, the light field depth information estimation algorithm for multi-occlusion interference handling is proposed. The occluder analysis of the angular patch image is carried out to analyze the position distribution characteristics of the occluder. Based on the classification idea, an improved Affinity Propagation（AP） clustering algorithm is proposed to classify the angular patch image, to separate the occluder from the scene point. The cost function combining the pixel intensity information entropy and the central variance is constructed on the angular patch image after separating the occluder. By minimizing the cost function, the initial depth estimation is obtained. The smooth constrained energy function based on MAP-MRF（Maximum A Posteriori Estimated Markov Random Field） framework is constructed and solved by the Graph Cut Algorithm（CCA） to obtain the final depth estimation. The experimental results show that compared with the existing depth estimation algorithm, the proposed algorithm improves the estimation accuracy at the occlusion.

Key words: light field, depth estimation, occlusion interference, image clustering, cost function, MAP-MRF framework

摘要：

针对光场的深度信息估计中，由遮挡带来的干扰，造成遮挡处的深度值估计精度低的问题，提出一种抗多遮挡物干扰的光场深度信息估计算法。对场景点的angular patch图像进行多遮挡物分析，分析遮挡物的位置分布特性。基于分类的思想提出改进AP（Affinity Propagation）聚类算法将场景点的angular patch图像进行像素点分类，将遮挡物和场景点分离。对分离遮挡物后的angular patch图像提出联合像素强度信息熵及中心方差的目标函数，最小化该函数，求得场景点的初始深度值估计。对初始深度值估计提出基于MAP-MRF（最大后验估计的马尔可夫随机场）框架的平滑约束能量函数进行平滑优化，并采用图割算法（Graph Cut Algorithm）求解，得到场景的最终深度值估计。实验结果表明，相较于现有深度信息估计算法，所提算法提升了遮挡处的估计精度。

关键词: 光场, 深度信息估计, 遮挡干扰, 图像聚类, 目标函数, MAP-MRF框架