基于共聚焦图像序列的深度估计方法

DEPTH ESTIMATION BASED ON CONFOCAL IMAGE SEQUENCE

提出一种基于共聚焦图像的深度估计方法。采用虚拟孔径技术把相机阵列获得的多视角图像合成得到共聚焦图像序列,并将其作为聚焦测距的数据源进行聚焦分析,实现对场景深度信息的估计。针对共聚焦图像的特点,将传统聚焦测距方法中的清晰度评价算法与颜色一致性评价算法相结合,提出一种聚焦度测量算法,该测量算法同时适用于图像中强纹理区域和弱纹理区域的聚焦分析。实验结果表明,该方法对聚焦判别的有效性较高,可以获得较准确的场景深度估计。

In this paper, a depth estimation approach based on confocal images is proposed. In order to realise the estimation of depth information of scenes, we use virtual aperture technique to set an array of cameras to acquire multi-angle images for obtaining confocal images sequence through synthesis and use it as the data resource of focus ranging to make focus analysis. Aiming at the feature of confocal images, we combines clarity assessing algorithm with colour constancy assessing algorithm of the conventional focus ranging method and put forward a focus degree metric algorithm, which suites for the focus analyses of both strong texture area and weak texture area in images. Experimental results show that this approach has higher effectiveness in focus judgment and can gain a quite accurate depth estimation of scenes.