Summary form only given. Traditional intra prediction methods for HEVC rely on using the nearest reference lines predicting a block, which ignore much richer context between current block and its neighboring blocks therefore cause inaccurate especially when weak spatial correlation exists lines. To overcome this problem, in paper, an intra-prediction convolutional neural network (IPCNN) is proposed prediction, exploits rich of capable improving accuracy block. Meanwhile, reconstruction three...

An ideal quality assessment model should simulate the properties of visual brain to be consistent with human evaluation. The appears have both evolved seek an efficient, decorrelated representation image information and "match" statistics natural image. On one hand, theoretical studies suggest that sparse resembles strategy in primary cortex for representing images. other scene driven evolution system also inspired understanding simulating perception. Inspired by these observations, this...

Full-reference (FR) image quality assessment (IQA) evaluates the visual of a distorted by measuring its perceptual difference with pristine-quality reference, and has been widely used in low-level vision tasks. Pairwise labeled data mean opinion score (MOS) are required training FR-IQA model, but is time-consuming cumbersome to collect. In contrast, unlabeled can be easily collected from an degradation or restoration process, making it encouraging exploit boost performance. Moreover, due...

360° videos have been widely used with the development of virtual reality technology and triggered a demand to determine most visually attractive objects in them, aka video saliency prediction (VSP). While generative models, i.e., variational autoencoders or autoregressive models proved their effectiveness handling spatio-temporal data, utilizing them VSP is still challenging due problem severe distortion feature alignment inconsistency. In this study, we propose novel consistency network...

Stereoscopic vision is a complex system which receives and integrates perceptual information from both monocular binocular cues. In this paper, novel reduced-reference stereoscopic image quality assessment scheme proposed, based on the visual measured by entropy of classified primitives (EoCP) mutual (MIoCP), named as DCprimary, sketch texture respectively, in accordance with hierarchical progressive process human perception. Specifically, EoCP each-view are calculated cue, MIoCP between...

In Virtual Reality (VR), the necessity of immersive videos leads to greater challenges in compression and communication owing much higher spatial resolution, rapid, often real-time changes viewing direction. Foveation displays exploits space-variant density retinal photoreceptors, which decreases exponentially with increasing eccentricity, reduce amount data from visual periphery. Foveated is gaining relevance popularity for Reality. Likewise, being able predict quality displayed foveated...

The human visual perception is a layered progressive process that brain assimilates information gradually, from primary information, structural to detailed information. Recently, the primitives (atoms in dictionary) extracted by sparse representation have been shown be highly related of perception. In this paper, are first classified into three categories: DCprimary, sketch and texture terms their inherent properties regarding tothe perceptual Then, we propose novel reduced reference (RR)...

The backward-compatible stereoscopic display is a technology that view perceived with 3D glasses while 2D version of the image concurrently available for naked-eye viewers on same physical medium. This unique functionality achieved by an information Temporal Psychovisual Modulation (TPVM), interesting interplay between high refresh rate optoelectronic display, signal processing and psychophysics. However, current performance system not satisfactory, it trade-off keeping simultaneously best...

As a promising solution for image retrieval in large-scale scenarios, hashing technique maps images into set of binary codes due to the storage and search efficiency. One recent trend is asymmetric deep supervised hash that could utilize information database. However, are still learned from local similarities. In this study, motivated by central similarity quantization, we present novel method takes both global similarities loss account. A new objective its efficient optimization proposed,...

<p>Obtaining large-scale noisy/clean image pairs from the real world to train a denoising model is difficult and cumbersome task. Self-supervised denoisers have gained recent attention by adopting blind-spot networks remove noise single noisy images. However, results of suffer losses detail caused selection random masks. Here, we propose an edge-enhanced approach called Self2Grad that compensates for loss important details when using self-supervised networks. Specifically, develop...

<p>Obtaining large-scale noisy/clean image pairs from the real world to train a denoising model is difficult and cumbersome task. Self-supervised denoisers have gained recent attention by adopting blind-spot networks remove noise single noisy images. However, results of suffer losses detail caused selection random masks. Here, we propose an edge-enhanced approach called Self2Grad that compensates for loss important details when using self-supervised networks. Specifically, develop...