Benchmarking of several disparity estimation algorithms for light field processing

Abstract

A number of high-quality depth imaged-based rendering (DIBR) pipelines have been developed to reconstruct a 3D scene from several images taken from known camera viewpoints. Due to the specific limitations of each technique, their output is prone to artifacts. Therefore the quality cannot be ensured. To improve the quality of the most critical and challenging image areas, an exhaustive comparison is required. In this paper, we consider three questions of benchmarking the quality performance of eight DIBR techniques on light fields: First, how does the density of original input views affect the quality of the rendered novel views? Second, how does disparity range between adjacent input views impact the quality? Third, how does each technique behave for different object properties? We compared and evaluated the results visually as well as quantitatively (PSNR, SSIM, AD, and VDP2). The results show some techniques outperform others in different disparity ranges. The results also indicate using more views not necessarily results in visually higher quality for all critical image areas. Finally we have shown a comparison for different scene's complexity such as non-Lambertian objects.

Attention Mechanism-Based Light-Field View Synthesis

“…In this section, we compare our proposed method with our previous work [25], namely LFVS-AM (Light-Field View Synthesis using Attention Module) and four state-of-theart light-field view synthesis methods utilizing four corner views. These methods include: Kalantari et al [22], Navarro et al [24], Jin et al [47], and DM-OTF [63]. Out of these four methods, DM-OTF (Direct Method Optical Flow) follows the traditional non-learning-based view synthesis pipeline using optical flow-based disparity estimation presented in Facebook Surround 360 [66].…”

“…Among the proposed variants, LFVS-AM-SAS attains the overall best results. LFVS-AM-SAS achieves a PSNR 5.65 dB better result than the traditional optical flow-based method DM-OTF [63] and more than 1 dB gain comparing to the best state-of-theart method, i.e., Navarro et al [24]. Due to the absence of an important depth cue (stereo) in the pipeline of Srinivasan et al [42], a significant degradation can be observed in its performance in contrast to the other schemes.…”

“…In Fig. 6, we compare the PSNR behaviour of individual views between all the methods except DM-OTF [63] and Srinivasan et al [42] because they perform the worst among all. It can be observed that our proposed method attains the highest PSNR values for all intermediate views, even for the farthest target position from all reference input views, i.e., the middle view.…”

A Benchmark of Light Field View Interpolation Methods

“…One of most common geometric representation is a depth map which can be used to synthesize novel views by warping the existing views. Traditional methods can infer depth maps from pairs of views using optical flow [21]. Wanner and Goldluecke [11] propose to use a variational method to synthesize novel view taking the inaccuracy caused by the depth estimations into consideration.…”

“…• DM-OTF: We also choose a traditional view synthesis pipeline using depth estimation and warping. Based on Zakeri et al [21], we choose the optical flow method used in Facebook Surround 360 [19] to obtain the disparity maps. To synthesis novel view, we backward warp the RGB information using disparity maps and holes are computed by the corresponding pixels from its neighbors.…”