Considering the problem that large amount of structural defects in the depth image captured by Kinect2 decreased the accuracy of object detection and recognition. This paper proposed a depth image enhancement algorithm that fused RGB image information. Firstly, a synchronized RGB image was introduced and aligned with the depth image. Secondly, based on the preprocessing of the color image and the depth image, an effective supporting edge region was extracted, the Just Noticeable Blur (JNB) threshold was used as the weight information of the depth map hole filling to repair the invalid area in the depth image. Finally, in order to highlight the details of edge features, we used an improved Guided Filter algorithm to enhance the edge details. Experimental results show that the algorithm can accurately fill the holes of the depth image and can effectively keep the edge details of the object.
Paper
Full text
Depth Image Enhancement Algorithm Based on RGB Image Fusion
Semantic Scholar · Computer Science · 2018
Abstract
Considering the problem that large amount of structural defects in the depth image captured by Kinect2 decreased the accuracy of object detection and recognition. This paper proposed a depth image enhancement algorithm that fused RGB image information. Firstly, a synchronized RGB image was introduced and aligned with the depth image. Secondly, based on the preprocessing of the color image and the depth image, an effective supporting edge region was extracted, the Just Noticeable Blur (JNB) threshold was used as the weight information of the depth map hole filling to repair the invalid area in the depth image. Finally, in order to highlight the details of edge features, we used an improved Guided Filter algorithm to enhance the edge details. Experimental results show that the algorithm can accurately fill the holes of the depth image and can effectively keep the edge details of the object.