Stereoscopic-3D (S3D) technology has gained widespread adoption recently across multimedia-based consumer applications due to its ability to develop immersive perceptual experiences. With this tremendous utilization, assessing the generated S3D content quality is essential for achieving consistent and reliable performance across diverse multimedia systems and devices. To automate this process, we develop a reference-free image quality assessment (IQA) algorithm based on performing a multiscale biorthogonal wavelet transform on the cohesive color map derived from an S3D image. We model the resulting sub-bands using a univariate generalized Gaussian distribution and estimate the corresponding fitting coefficients. We demonstrate that the computed features effectively distinguish distortions and estimate the Chi-square distance between test image features and pristine model parameters to measure the primitive chrominance feature of an S3D image. Further, we perform a second derivative Laplacian operator on the cohesive color map to measure the overall sharpness of chromatic information of an S3D scene. Next, we compute the perceptual image quality evaluator on the left and right scenes to estimate the overall luminance feature of an S3D image. Finally, chromatic and luminance-based features are linearly combined to calculate the overall quality score of an S3D image. We evaluate our model on four benchmark datasets (LIVE Phase I & II, Waterloo Phase I & II), demonstrating its superior performance against eighteen existing IQA models, including 2D and 3D opinion-unaware and aware algorithms. This algorithm provides a highly effective solution to ensure the data quality required for the next generation of multimedia applications. The source code for the proposed opinion-unaware model can be accessed via the following: Google Drive link.

Stereoscopic (S3D) images, as an advanced visual multimedia format, are gaining popularity among consumers and researchers due to their ability to provide immersive and realistic experiences compared to conventional 2-D content. However, similar to the 2-D imaging pipeline, S3D images are susceptible to quality degradation and perceptual losses during acquisition, encoding, transmission, and display systems. Therefore, automatic quality assessment (QA) of S3D images without pristine S3D features and training/testing mechanisms is of utmost importance. In this work, we propose an unsupervised no-reference (NR) S3D image QA (IQA) model using a cohesive color map generated from the individual views of an S3D image. The quality-aware representations of the cohesive map are computed by partitioning it into nonoverlapping blocks, followed by applying singular value decomposition (SVD) to each block and utilizing its singular values and eigenvector matrices to capture the distortion perturbations of S3D images. Furthermore, the quality score of the cohesive map is computed by averaging the harmonic distance between the multivariate Gaussian (MVG) parameter of pristine images and the corresponding block-level features of a test S3D image. Finally, a weighted parametric pooling function is employed to fuse the cohesive map quality score with spatial quality scores derived from the PIQE algorithm to obtain the final perceptual quality score of a test S3D image. The superiority of our proposed S3D QA model has been demonstrated through extensive experiments involving 16 off-the-shelf models across four benchmark S3D IQA datasets. The proposed model outperformed three opinion-unaware 2-D NR IQA models and three S3D NR IQA approaches and also delivered competitive performance against ten opinion-aware S3D NR IQA models. Notably, this performance was achieved without relying on any regression or deep learning modules, highlighting the model’s robustness in estimating the perceptual quality of S3D images.

This letter presents an automatic and opinion-unaware quality assessment (QA) model for stereoscopic (3-D) images affected by sensor, transmission, and compression distortions. The proposed QA model does not use reference 3-D images or subjective ratings from the primary dataset to map to a quality score. This design is important in 3-D image quality assessment (IQA), where generalization across various 3-D IQA datasets has remained a major challenge. Therefore, to address this gap, we propose a QA model for 3-D images by computing a correlation map between the luminance and depth components. Subsequently, we employ a univariate generalized Gaussian distribution (UGGD) to model the multiscale and multiorientation steerable decomposed subbands and capture the quality-aware representations of the computed correlation map. Then, the spread and the shape parameters of UGGD of the test 3-D image are estimated, and the individual likelihood estimates are calculated from the pristine multivariate Gaussian parameters of the corresponding shape and spread features. Experiments on LIVE Phase I and Phase II datasets show that the proposed opinion-unaware model outperforms existing opinion-unaware models on Phase I and achieves competitive performance against opinion-aware models. On Phase II dataset, it delivers competitive results compared to opinion-aware and opinion-unaware approaches.

The rapid growth of stereoscopic (S3D) imaging has led to increased adoption across multimedia formats. However, the S3D pipeline introduces spatial, depth, and binocular rivalry errors, degrading visual quality. Quality assessment (QA) helps evaluate and improve the visual experience, ensuring comfort during prolonged exposure to S3D content. This article proposes an "opinion-unaware"QA model for S3D images based on statistical variations in the correlation map between color and depth features. The quality score of the correlation map is obtained using principal component analysis (PCA) and multivariate Gaussian modeling. A natural image quality evaluator calculates spatial scores from the luminance information of an S3D image. Finally, the overall S3D image quality is determined by a weighted mechanism that combines color, depth, and spatial scores. The weighted pooling includes the spatial score, which is otherwise ignored in correlation map calculation. Extensive experiments on four standard datasets, namely, LIVE Phase I and Phase II, Waterloo Phase I and Phase II, have demonstrated that the proposed model outperforms off-the-shelf opinion-unaware models and gives competitive performance to supervised no-reference (NR) models.