Recent advances in vision modeling for image and video processing

Computational models of the human visual system (HVS) play an instrumental role in the design and operation of numerous image and video processing algorithms. HVS models have proved beneficial not only for boosting an algorithm’s performance, but also for improving its reliability and robustness to changing conditions. Many processing algorithms operate on images and videos that are meant to be viewed by humans. For such algorithms, HVS models can help on multiple fronts. First, many HVS models can be directly incorporated into one or more stages of a processing algorithm. This approach is quite beneficial because it allows an algorithm to take into account aspects of visual perception that would otherwise be neglected. Second, even if an HVS model is too time-consuming or resource-intensive to be integrated into an algorithm, the model can still be useful for guiding an algorithm’s design. For example, an HVS model can be used to evaluate the perceptual impacts of a particular processing stage, which then allows the engineer to make more perceptually optimal design decisions, and thereby allows the engineer to indirectly incorporate aspects of visual perception into an algorithm. A third benefit of HVS modeling is the possibility to learn from the processing strategies employed in the models. Many effective analysis techniques such as sparse coding, divisive normalization, and adaptive feature selection have proven crucial for accurately modeling psychophysical results. Subsequently, these approaches have found widespread use in a variety of algorithms such as coding, segmentation, and quality assessment. In this way, an HVS model can serve as an intermediary for transferring biological processing strategies employed by the HVS into computational processing strategies employed by an algorithm. This special issue represents a diverse set of research efforts on recent advances in HVS modeling for image and video processing—including psychophysical experiments, improved perceptual and saliency modeling, quality assessment, and novel applications of HVS models. Ultimately, all