Improved depth map upsampling using iterative joint trilateral filter

Abstract

Typical depth sensor cameras have a lower resolution compared to RGB cameras. In this work, inspired by the Joint Trilateral Filter (JTF), an iterative JTF (JTFite) approach is presented to improve the visual quality of the depth discontinuity regions. Our framework takes into consideration the utilization of the difference of refined pixels' values and the canny edge of the JTF, to obtain sharper edges. The recent work on JTF is revisited by setting varying parameters which employs local gradients information to avoid texture artifacts in a structure-aware onion ltering. We extend the JTF by using traits of its local ltering technique to improve further the edges of the objects in the depth map by post-ltering the updated region. The proposed method enhances details of the textures in the depth map by emphasizing on gradients of the drastic changes of the pixels values. The primary goal of this work is to increase the depth resolution by a high-resolution RGB image to preserve sharp edges and details of the textures. Visual and quantitative experimental results show that the proposed method outperforms the JTF for depth map upsampling. We evaluate the upsampled and ltered low-resolution (LR) depth maps with their associated ground truth (GT) images.