Perception-Oriented Stereo Image Super-Resolution

Abstract

Recent studies of deep learning based stereo image super-resolution (StereoSR) have promoted the development of StereoSR. However, existing StereoSR models mainly concentrate on improving quantitative evaluation metrics and neglect the visual quality of superresolved stereo images. To improve the perceptual performance, this paper proposes the first perception-oriented stereo image superresolution approach by exploiting the feedback, provided by the evaluation on the perceptual quality of StereoSR results. To provide accurate guidance for the StereoSR model, we develop the first special stereo image super-resolution quality assessment (StereoSRQA) model, and further construct a StereoSRQA database. Extensive experiments demonstrate that our StereoSR approach significantly improves the perceptual quality and enhances the reliability of stereo images for disparity estimation. CCS CONCEPTS• Computing methodologies → Reconstruction.

NTIRE 2022 Challenge on Stereo Image Super-Resolution: Methods and Results

“…More recently, Wang et al [33] modified PAM [27] to be bidirectional and symmetric, and developed an improved version of PASSRnet (i.e., iPASSR) to handle a series of practical issues (e.g., illuminance variation and occlusions) in stereo image SR. Dai et al [34] proposed a feedback network to alternately solve disparity estimation and stereo image SR in a recurrent manner. Ma et al [35] proposed a GAN-based perception-oriented stereo image SR method that can generate visually pleasing and stereo consistent details. Xu et al [36] tackled the stereo video SR problem by simultaneously utilizing both cross-view and temporal information.…”