A Bidirectional Guided Filter Used for RGB-D Maps

The red-green-blue-depth (RGB-D) cameras are widely used in the fields of robotic spray painting, packaging, and tracking. However, the quality of the depth maps of RGB-D maps is relatively low. Guided filters utilizing the color image as a guide map is a common method to deal with these problems. However, the noise in guidance will make the filtering output contains unexpected noise that does not contain before filtering. In this article, based on the classical guided filter, a bidirectional guided filter is proposed to filter the filtering input and the guidance simultaneously. The basic equations are proposed and the filter kernel is derived. The properties of the proposed bidirectional guided filter, such as edge-preserving filtering property, insensitivity to the noise of guidance, and filtering effects on guidance, are investigated. Four different groups of experiments are performed to further verify the effectiveness of the proposed method. The classical guided filter and the bidirectional guided filter are compared to verify that the proposed bidirectional guided filter can reduce the impact of the noise of the guidance; and the bidirectional guided filter and the box mean filter are compared to verify the filtering effect on guidance of the proposed bidirectional guided filter.