Multi–Scale Synthesized View Assessment Based on Morphological Pyramids

Abstract

The Depth-Image-Based-Rendering (DIBR) algorithms used for 3D video applications introduce geometric distortions affecting the edge coherency in the synthesized images. In order to better deal with specific geometric distortions in the DIBR synthesized images, we propose full-reference metric based on multi-scale pyramid decompositions using morphological filters. The non-linear morphological filters used in multi-scale image decompositions maintain important geometric information such as edges across different resolution levels. We show that PSNR has particularly good agreement with human judgment when it is calculated between detailed images at higher scales of morphological pyramids. Consequently, we propose reduced morphological pyramid peak signal-to-noise ratio metric (MP-PSNR), taking into account only mean squared errors between pyramids' images at higher scales. Proposed computationally efficient metric achieves significantly higher correlation with human judgment compared to the state-of-the-art image quality assessment metrics and compared to the tested metric dedicated to synthesis-related artifacts.

No-Reference Light Field Image Quality Assessment Based on Micro-Lens Image

“…We conducted comparison experiments between our proposed model and several FR, RR and NR metrics, including nine 2D-FR metrics [10][11][12][28][29][30][31][32], three 2D-NR metrics [15,33,34], one 3D-FR metric [35], two 3D-NR metrics [16,36], five Multi-views FR metrics [13,14,37,38], one Multi-views NR metric [39] and one LF-RR metric [18]. Three evaluation criteria are selected to measure the correlation between MOS and predicted results, consisting of Spearman Rank Order Correlation Coefficient (SROCC), Linear Correlation Coefficient (LCC) and Root Mean Squared Error (RMSE).…”

“…MS-SSIM [11] and FSIM [12]. MP-PSNR Full [13] and MP-PSNR Reduc [14] based on Morphological pyramid decomposition are proposed to evaluate the multi-view image quality. RR methods only require part of information from original images.…”

Towards a Quality Metric for Dense Light Fields

“…Battisti et al [2] proposed more sophisticated framework for evaluating the quality of depth image based rendering techniques by comparing the statistical features of wavelet subbands and used image registration and skin detection steps for additional optimization. Sandic et al [30] exploited multi-scale pyramid decompositions with morphological filters for obtaining the quality of intermediate views and showed that they achieve significantly higher correlation with subjective scores. These methods form a class of metrics specific to view-interpolation artifacts, and 2D stimuli containing the interpolated views are used for subjective experiments.…”

DIBR-Synthesized Image Quality Assessment via Statistics of Edge Intensity and Orientation

“…RMP-PSNR [7] and 3DSwIM [8]. Table 1 summarizes the experiment results, where the best results are marked in boldface.…”

“…In [6], MW-PSNR was improved by using morphological pyramids, producing the Morphological Pyramid Peak Signalto-Noise Ratio (MP-PSNR) method. The reduced version of MP-PSNR (RMP-PSNR) [7] was also presented, which only used MSEs on higher scales. In [8], the blocks in reference and synthesized images were first registered.…”