Multimedia data has the characteristics of large scale and skewed distribution with a long-tailed shape, which is challenging imbalance problem faced by deep learning. In image instance segmentation, existing methods deal this from single perspective, ignoring presence multiple factors, results in limitation performance. Considering that imbalances exist not only between positive negative classes, but also foreground background subclasses, as well hard easy examples, we argue losses samples...

Human pose estimation from image or video is a basic issue in computer graphics and vision. The challenge of human lies the temporal coherency issue. consistency contents' similarity shown frames. In video, maintenance to obtain better long-term consistency. Great major methods for are using whole optimization method, which makes very large computation absence before after articulated limbs. this paper, novel method proposed. We maintain by structured space learning halfway evaluation...

Photographs taken by mobile device usually suffer from loss of details and low visual attraction due to the poor light condition. The enhancement underexposed image can effectively solve this problem. However, previous work may inevitably wash out some weak edges lose when handling several images. To deal with these problems, paper presents a detail-preserving method based on new optimal weighted multi-exposure fusion mechanism. Providing an input image, we propose novel which generate...

Numerous screen content images (SCIs) have been produced to meet the needs of virtual desktop and remote display, which put forward a very urgent requirement for security management SCIs. Perceptual hashing is an effective way deal with this issue. However, since SCIs are generally composed pictures, graphics texts, their intrinsic characteristics different from those natural images. Thus previous methods not suitable In article, we propose perceptual method perspective visual understanding....

Content-based image copy detection has become one of the important technologies in copyright protection, where two major processes, content-based feature extraction and matching are included. However, it is certainly true that enough storage space required to establish database for matching, which greatly increases time consumption, as well lacks flexibility. Fortunately, perceptual hashing a good strategy address these problems, features extracted further encoded hash codes. On hand,...

Recently, neural style transfer has become a popular task in both academic research and industrial applications. Although the existing methods made great progress terms of quality efficiency, most them mainly focus on extracting high-level features. Therefore, it is still challenging to display hierarchical structure content image due lack texture information, which causes blurred boundaries distortion stylized image. In this paper, novel video scheme proposed suppress preserve semantic...

Nowadays the problem of image quality assessment (IQA) for screen content images (SCIs) has become a research hotspot as they are ubiquitous in multimedia applications. Although natural (NIs) been continuously developed past few decades, NI-oriented IQA methods can be directly applied on SCIs due to different visual characteristics between them. In this paper, we present no-reference prediction approach considering information SCIs, which is based dual-channel multi-task convolutional neural...

Traditional color transfer methods can achieve satisfactory results for transferring the style from a reference image to source image, provided that shares similar mood with image. However, solutions are always sensitive category, which cannot generate natural when contents of and different, e.g., lush tree in bare In this situation, it is insufficient only through appearance transfer, since other information such as texture should also be considered. To obtain sufficient results, we propose...

A typical rainfall scenario contains tens of thousands dynamic sound sources. characteristic the large-scale scene is strong randomness in raindrop distribution, which makes it notoriously expensive to synthesize such sounds with purely physical methods. Moreover, raindrops hitting different surfaces (liquid or various solids) can emit distinct sounds, for prior methods unified impact models are ill-suited. In this paper, we present a physically-based statistical simulation method realistic...

The main purpose of infrared and visible image fusion is to produce a that incorporates less redundant information while incorporating more complementary information, thereby facilitating subsequent high-level visual tasks. However, obtaining from different modalities images challenge. Existing methods often consider only relevance neglect the complementarity modalities’ features, leading loss some cross-modal information. To enhance it believed comprehensive interactions should be provided....

Hashing method is an efficient technique of multimedia security for content protection. It maps image into a content-based compact code denoting the itself. While most existing algorithms focus on improving classification between robustness and discrimination, little attention has been paid to geometric invariance under normal digital operations, therefore results in quite fragile distortion when applied copy detection. In this article, novel effective hashing proposed based invariant vector...

Image decolorization is a task aiming to transform color image grayscale one and dimension reduction process which inevitably suffers from information loss. The general goal of preserve the contrast image. According human visual study, exposure affects perception, low-exposure areas or over-exposure will first attract sense sight. In addition, also image, often cannot be well shown. Thus, should taken into account in decolorization. Traditional local methods are not accurate enough pixel...

In panoramic multimedia applications, the perception quality of omnidirectional content often comes from observer's viewports and overall impression after browsing. Starting this hypothesis, paper proposes a deep-learning based joint network to model no-reference assessment images. On one hand, motivated by different scenarios that lead human understandings, convolutional neural (CNN) is devised simultaneously encode local features latent rules viewports, which are more likely be noticed...