No-Reference Light Field Image Quality Assessment Based on Spatial-Angular Measurement

Light field image quality assessment (LFI-QA) is a significant and challenging research problem. It helps to better guide light field acquisition, processing and applications. However, only a few objective models have been proposed and none of them completely consider intrinsic factors affecting the LFI quality. In this paper, we propose a No-Reference Light Field image Quality Assessment (NR-LFQA) scheme, where the main idea is to quantify the LFI quality degradation through evaluating the spatial quality and angular consistency. We first measure the spatial quality deterioration by capturing the naturalness distribution of the light field cyclopean image array, which is formed when human observes the LFI. Then, as a transformed representation of LFI, the Epipolar Plane Image (EPI) contains the slopes of lines and involves the angular information. Therefore, EPI is utilized to extract the global and local features from LFI to measure angular consistency degradation. Specifically, the distribution of gradient direction map of EPI is proposed to measure the global angular consistency distortion in the LFI. We further propose the weighted local binary pattern to capture the characteristics of local angular consistency degradation. Extensive experimental results on four publicly available LFI quality datasets demonstrate that the proposed method outperforms state-of-the-art 2D, 3D, multi-view, and LFI quality assessment algorithms.

💡 Research Summary

Light‑field (LF) imaging captures both spatial intensity and angular direction of light rays, offering rich information for applications such as refocusing, depth estimation, and immersive display. However, assessing LF image quality is challenging because degradation can affect spatial fidelity of individual sub‑aperture images (SAIs) and the angular consistency among them. Existing objective quality metrics either ignore angular information (2D IQA), rely on reference data (full‑reference methods), or are designed for multi‑view/3D content without fully exploiting LF characteristics.

This paper introduces the first no‑reference LF quality assessment framework (NR‑LFQA) that jointly evaluates spatial quality and angular consistency. The spatial component is modeled through a Light‑field Cyclopean Image Array (LFCIA), which simulates binocular fusion and rivalry by combining pairs of SAIs into a set of cyclopean images. The naturalness distribution of LFCIA is quantified using a statistical model derived from natural scene statistics, yielding the Light‑field Cyclopean image Naturalness (LCN) feature. LCN captures global spatial degradations such as blur, compression artifacts, and color shifts across all SAIs.

For angular consistency, the method leverages Epipolar Plane Images (EPIs), 2‑D slices obtained by fixing one spatial coordinate and varying the angular coordinate. Two complementary descriptors are extracted: (1) Gradient Direction Distribution (GDD), which computes the histogram of gradient orientations on EPIs to reflect global disruption of the linear structures that encode angular coherence; (2) Weighted Local Binary Pattern (WLBP), an extension of the classic LBP that assigns angular‑dependent weights to neighboring pixels, thereby detecting local inconsistencies caused by interpolation errors or compression artifacts.

The three feature vectors (LCN, GDD, WLBP) are concatenated and fed into a regression model (multi‑linear regression or SVR) trained on subjective scores. The authors evaluate NR‑LFQA on four publicly available LF quality datasets: WIN5‑LID, MPI‑LFA, SMART, and VALID. These datasets contain a mixture of real and synthetic LFIs, multiple distortion types (HEVC, JPEG2000, linear/nearest‑neighbor interpolation, CNN‑based artifacts, etc.), and subjective ratings expressed as MOS, JOD, or Bradley‑Terry scores.

Across all datasets, NR‑LFQA achieves higher Pearson and Spearman correlation coefficients and lower RMSE than state‑of‑the‑art 2D FR/IQA (SSIM, MS‑SSIM, BRISQUE, NIQE), 3D FR/IQA (3DSwIM, SINQ), multi‑view NR/IQA (APT, MP‑PSNR), and recent LF‑specific FR methods (gradient magnitude similarity on EPIs). Ablation studies confirm that both the spatial LCN and the angular GDD/WLBP components contribute significantly; removing either leads to noticeable performance drops, especially for distortions that primarily affect angular coherence.

In summary, the paper makes two major contributions: (1) a novel cyclopean‑image‑based naturalness metric that aggregates spatial quality across the entire LF, and (2) an EPI‑based dual descriptor (global gradient direction and weighted local binary patterns) that accurately captures angular consistency degradation. The proposed NR‑LFQA operates without any reference LF, making it suitable for real‑time monitoring and automatic optimization in LF acquisition, compression, and display pipelines. Future work may explore deeper learning models for feature extraction, extension to high‑dynamic‑range LF, and computationally efficient implementations for embedded devices.