How to explore the interaction between RGB and thermal modalities is key success of RGB-T saliency object detection (SOD). Most existing methods integrate multi-modality information by designing various fusion strategies. However, modality gap features will lead unsatisfactory performances simple feature concatenation. To solve this problem, we innovatively propose a cross-guided difference reduction network (CGMDRNet) achieve intrinsic consistency via reducing differences. Specifically,...

The ability of capturing the complementary information multi-modality data is critical to development salient object detection (SOD). Most existing studies attempt integrate through various fusion strategies. However, most these methods ignore inherent differences in data, resulting poor performance when dealing with some challenging scenarios. In this paper, we propose a novel Modality-Induced Transfer-Fusion Network (MITF-Net) for RGB-D and RGB-T SOD by fully exploring complementarity...

RGB-T salient object detection (SOD) aims to detect and segment saliency regions on RGB images the corresponding thermal maps. The ability of alleviating modality difference between plays a vital role in development SOD. However, most existing methods try integrate multi-modal information through various fusion strategies, or reduce via unidirectional undifferentiated bidirectional interaction, but failing some challenging scenes. To deal with above question, novel Cross-Modality Double...

Vision-based measurement techniques are required in the quality inspection process of various products. However, most existing research methods focus on use a single modality (RGB image or Depth map) for defect detection. In this paper, we propose potential detection technique by introducing RGB-D salient object (SOD) as method and presenting Hierarchical Fusion Multi-Level Decoder Network (HFMDNet). The key to recently popular multi-modal SOD lies effectively acquiring cross-modal...

Spatio-temporal-spectral fusion aims to produce high spatio-temporal-spectral resolution images by integrating the complementary spatial, temporal, and spectral advantages of multi-source remote sensing images. However, on one hand, existing methods are insufficient exploit inherent complex nonlinear relationship among multisource multitemporal observations. On other since unavailability real images, it is difficult adopt deep learning with supervised training. In this paper, we propose an...

In reviewing the research progress in Point Cloud Quality Assessment (PCQA), two main pathways have emerged, i.e., 2D projections and 3D point descriptors. The former primarily focuses on visual information, while latter concentrates crucial geometrical information three-dimensional space. However, current studies lack a thorough investigation of impact components seldom pay special attention to plane-point fusion strategies. To comprehensively represent features effectively tackle various...

The aesthetic and appreciation of an image is the innate human perceptual ability. Emotion, as one most basic perceptions, has been found to have a close relationship with aesthetics. However, explicitly incorporating learned emotion cues into aesthetics assessment (IAA) model remains challenging. Additionally, humans consider both fine-grained details holistic context information in assessments. Therefore, utilization emotional enhance modulate representation features detail crucial for...

The challenge of detecting helmet-wearing on non-motorized vehicles within road traffic scenarios has long been beset by issues like inadequate feature extraction and background noise interference. To address these challenges, an algorithm tailored for amidst complex environments was proposed in this paper. This employs enhancement techniques context-aware fusion strategies to effectively the considerable challenges presented vast quantity vehicles, small target dimensions, need accurate...

In the animation industry, automatically predicting quality of cartoon images based on inputs general distortions and color change is an urgent task, while existing no-reference (NR) methods usually measure perceptual natural images. this paper, observation that structure are main factors affecting quality, we proposed a new NR prediction metric for images, which fully takes gradient information into account. The experimental results our newly constructed NBU-CIQAD dataset with other image...

Arbitrary neural style transfer is a vital topic with great research value and wide industrial application, which strives to render the structure of one image using another. Recent researches have devoted efforts on task arbitrary (AST) for improving stylization quality. However, there are very few explorations about quality evaluation AST images, even it can potentially guide design different algorithms. In this paper, we first construct new images assessment database (AST-IQAD), consists...

Although the research of arbitrary style transfer (AST) has achieved great progress in recent years, few studies pay special attention to perceptual evaluation AST images that are usually influenced by complicated factors, such as structure-preserving, similarity, and overall vision (OV). Existing methods rely on elaborately designed hand-crafted features obtain quality factors apply a rough pooling strategy evaluate final quality. However, importance weights between will lead unsatisfactory...

Stereoscopic OmniDirectional Images (SODIs) usually require recording very High-Resolution (HR) information whereby it is beneficial to exploit a Super-Resolution (SR) scheme super-resolve Low-Resolution (LR) SODIs. Compared with traditional 2D SR approaches, the algorithms of SODI (SODI-SR) need deal two extra aspects: binocular and panoramic characteristics. In this paper, we first build synthetic-specific SODI-SR dataset LR-HR image pairs. Then propose Dynamic convolutions Transformer...

Stitched omnidirectional images are created by capturing overlapping and seamlessly merging them together. When observing stitched images, users wear head-mounted display that encompass their entire visual field positioned near the eyes. The presence of significant distortion in presented scene can lead to pronounced physical mental discomfort. Therefore, it becomes crucial explore methods for effectively evaluating quality images. Given process deviates from traditional resembling more...

How to explore the interaction between image aesthetic rules and crops is key finding views with good composition. Besides, it subjective evaluate candidate crops, which mainly depends on knowledge, but not an easy task for people without extensive photography experience. However, existing methods mostly find by extracting general features of fully exploring rules. Motivated this, we innovatively propose a composition-guided cropping assessment network (CGICAANet) efficiently optimizing...