This paper proposes a novel deep learning framework named bidirectional-convolutional long short term memory (Bi-CLSTM) network to automatically learn the spectral-spatial features from hyperspectral images (HSIs). In network, issue of spectral feature extraction is considered as sequence problem, and recurrent connection operator across domain used address it. Meanwhile, inspired widely convolutional neural (CNN), convolution spatial incorporated into extract feature. addition, sufficiently...

In this paper, we present a new method for facial age estimation based on ordinal discriminative feature learning. Considering the temporally and continuous characteristic of aging process, proposed not only aims at preserving local manifold structure images, but also it wants to keep information among faces. Moreover, try remove redundant from both locality as much possible by minimizing nonlinear correlation rank correlation. Finally, formulate these two issues into unified optimization...

Human motion capture data has been widely used in many areas, but it involves a complex process and the captured inevitably contains missing due to occlusions caused by actor's body or clothing. Motion recovery, which aims recover underlying complete sequence from its degraded observation, still remains as challenging task nonlinear structure kinematics property embedded data. Low-rank matrix completion-based methods have shown promising performance short-time-missing recovery problems....

Siamese networks have been successfully introduced into visual tracking, which match the best candidate and a target template via couple of with shared parameters. However, most network-based trackers (SNTs) are tailored to canonical posture search-region images, resulting in inferior performance when objects large-scale pose variations. Besides, SNTs fail discriminate distractors well because they only leverage high-level semantic features as representations that cannot tell from different...

Multimodal sentiment analysis (MSA) identifies individuals' states in videos by integrating visual, audio, and text modalities. Despite progress existing methods, the inherent modality heterogeneity limits effective capture of interactive features across In this paper, introducing a Multi-Modality Collaborative Learning (MMCL) framework, we facilitate cross-modal interactions enhanced complementary from modality-common modality-specific representations, respectively. Specifically, design...

Soil moisture (SM) plays an important role in hydrological cycle and weather forecasting. Satellite provides the only viable approach to regularly observe large-scale SM dynamics. Conventionally, is estimated from satellite observations based on radiative transfer theory. Recent studies have demonstrated that neural network (NN) method can retrieve with comparable accuracy as conventional methods. Here, we are interested whether NN model more complex structures, namely deep convolutional...

The fusion of spatial and spectral information in hyperspectral images (HSIs) is useful for improving the classification accuracy. However, this approach usually results features higher dimension curse dimensionality problem may arise resulting from small ratio between number training samples features. To ease problem, we propose a novel algorithm spatial-spectral feature extraction based on hypergraph embedding. Firstly, each HSI pixel regarded as vertex joint extended morphological...

Psychological and cognitive findings indicate that human visual perception is attentive selective, which may process spatial appearance selective attentions in parallel. By reflecting some aspects of these attentions, this paper presents a novel correlation filter (CF)-based tracking approach, corresponding to processing local semi-local background domains, respectively. In the domain, inspired by Gestalt principle figure-ground segregation, we leverage an efficient Boolean map...

By exploiting the kernel trick, sparse subspace model is extended to nonlinear version with one or a combination of predefined kernels, but high-dimensional space induced by kernels not guaranteed be able capture features data in theory. In this article, we propose nonconvex low-rank learning framework an unsupervised way learn replace model. The learned relaxation rank can better property induce Hilbert that more closely approaches true feature space. Furthermore, give global closed-form...

Motion synthesis technology can produce natural and coordinated motion data without a capture process, which is complex costly. Current methods usually provide few interfaces to avoid the arbitrariness of but this actually reduces understandability process. In paper, we propose learning-based Sphere nonlinear interpolation (Snerp) model that generate in-between motions in terms given start-end frame pair. Variety input pairs will enrich diversity generated motions. The angle speed human not...

Recently, a simple, yet effective and efficient tracker named Staple has achieved promising performance in terms of efficiency accuracy on series visual tracking benchmarks. is equipped with complementary learners discriminative correlation filters (DCFs) color histograms, which are robust to both changes deformations. However, it some drawbacks: 1) only employs standard histograms the same quantization step for all sequences, does not consider specific structural information target each...

In this paper, we present a new scheme for face recognition. The main idea is to represent the images with similarity features against reference set and provide relative match two images. For any image, first compute similarities between it all images, then take these as its feature. Based on features, linear discriminating classifier constructed recognize querying image. Inspired by research in cognitive psychology, perceptual distance based dynamic function proposed features. method can be...

In this paper, we present an effective hierarchical shot classification scheme for broadcast soccer video. We first partition a video into replay and non-replay shots with logo detection. Then, are further classified Long, Medium, Close-up or Out-field types color texture features based on decision tree. tested the method real FIFA videos, experimental results demonstrate its effectiveness..