Search

This paper presents a novel classification method based on spatial–spectral low-rank representation in the hidden field under a Bayesian framework for hyperspectral imagery. The key idea of the method is to simultaneously explore the low-rank property in the spectral domain and nonlocal self-similarity in the spatial domain of the hidden field, which is estimated by sparse multinomial logistic regression in a supervised manner. First, the low rank property in the spectral domain is exploited in local cubic patches. Following this, similar cubic patches are clustered into several groups in a nonlocal sense and patches in each group are assumed to lie in a low-rank subspace. The final model could be efficiently solved by the augmented Lagrangian method. Experimental results on two real hyperspectral datasets validate that the proposed classifier produces a superior performance compared to other state-of-the-art classifiers in terms of overall accuracy, average accuracy and the kappa statistic (k).

The rate-distortion optimized quantization (RDOQ) in HEVC has improved the coding efficiency of the conventional uniform scalar quantization (SQ) very much. Since the RDOQ is computationally complex, in this paper, we investigate a way of performing RDOQ more efficiently in HEVC. Based on our statistical observation of non-trivial percentage of transform blocks (TB) for which RDOQ does not change their quantization results of SQ, we design a learning-based quantizer selection scheme which can tell in advance whether RDOQ is expected to modify the quantization levels calculated by SQ. Only those TBs likely to be changed by RDOQ are subject to the actual RDOQ process. For the remaining TBs, we design an improved SQ which adapts the dead-zone interval size and round offset based on coefficient group and entropy coding features. The proposed improved SQ has much lower computational complexity than RDOQ while achieving better coding efficiency than the conventional SQ. The experimental results show that our efficient quantization scheme respectively provides 9% and 34% of encoding and quantization time reduction by selectively performing RDOQ only for 21% of TBs. The average BDBR performances of Y, Cb, and Cr channels are respectively–0.03%, 0.48%, and 0.45%.

One of the design goals of the recently published international video coding standard, Versatile Video Coding (VVC/H.266), is efficient coding of computer-generated video content (commonly referred to as screen content) which exhibits different signal characteristics from the usual camera-captured video (commonly referred as natural content). VVC can perform transform in multiple different ways including skipping the transform itself, which demands much computation for its best selection among many combinatory options. In this paper, we investigate designing a machine-learning-based early transform skip mode decision (ML-TSM) which makes a determination whether or not to skip the transform in an early stage by making a simple classification employing key features designed in such a way to reflect the characteristics of TSM blocks well. Compared with the VVC reference software 14.0, the proposed scheme is verified to reduce computational complexity by 11% and 4% with a Bjøntegaard delta bitrate (BDBR) increase of 0.34% and 0.23% respectively under all-intra (AI) and random-access (RA) configurations.

The dependent quantization (DQ) technique in Versatile Video Coding (VVC) provides high compression efficiency, but it suffers from much computation, which is not preferred in practical applications. In this context, we propose a complexity-efficient dependent quantization structure employing a machine learning-based DQ start position prediction and a simplified trellis structure. Firstly, the proposed DQ start position is decided by a binary classifier which finds an appropriate location to start the DQ process. To further reduce the processing time of the binary classification, we also set up a precondition with a learned preset threshold which can reduce the number of coefficients subjected to the classifier. Based on our statistical analysis showing that under certain conditions some branches of the trellis structure are not necessary to check, we also design a branch pruning method and a removal scheme of all-zero coefficient group checking to simplify the DQ trellis structure. Our proposed complexity-efficient DQ has much fewer calculations of rate-distortion costs. Our experimental results show that it can reduce the computational complexity of the DQ process with nearly no performance loss of Bjøntegaard delta bit rate (BDBR) irrespective of the all intra or the random access configurations.

The joint use of multisource remote-sensing (RS) data for Earth observation missions has drawn much attention. Although the fusion of several data sources can improve the accuracy of land-cover identification, many technical obstacles, such as disparate data structures, irrelevant physical characteristics, and a lack of training data, exist. In this article, a novel dual-branch method, consisting of a hierarchical convolutional neural network (CNN) and a transformer network, is proposed for fusing multisource heterogeneous information and improving joint classification performance. First, by combining the CNN with a transformer, the proposed dual-branch network can significantly capture and learn spectral–spatial features from hyperspectral image (HSI) data and elevation features from light detection and ranging (LiDAR) data. Then, to fuse these two sets of data features, a cross-token attention (CTA) fusion encoder is designed in a specialty. The well-designed deep hierarchical architecture takes full advantage of the powerful spatial context information extraction ability of the CNN and the strong long-range dependency modeling ability of the transformer network based on the self-attention (SA) mechanism. Four standard datasets are used in experiments to verify the effectiveness of the approach. The experimental results reveal that the proposed framework can perform noticeably better than state-of-the-art methods. The source code of the proposed method will be available publicly at https://github.com/zgr6010/Fusion_HCT.git.

Despite the remarkable progress made by the salient object detection of natural sensing images (NSI-SOD), the complex background and scale diversity issues of remote sensing images (RSIs) still pose a substantial obstacle. In this study, we build an end-to-end channel-enhanced remodeling-based network (CRNet) for optical RSIs (ORSIs) to highlight salient objects through feature augmentation. First, the backbone convolutional block is used to suggest the fundamental characteristics. Then, we use the channel enhance module (CEM) to enhance the shallow features. CEM primarily relies on the channel attention (CA) mechanism and uses a no-downscaling strategy to produce local cross-channel interaction, which lowers model complexity while enhancing extraction performance. Meanwhile, we use the redefined feature module (RFM) to reconstruct the deep features and generate global attention features by dimensional transformation and feature relationship aggregation to achieve the role of locating salient targets. Finally, the cascade combines the multiscale features to provide the final saliency map. To further enhance the representational power of the network, we use a hybrid loss function to improve performance. The proposed approach outperforms current state-of-the-art (SOTA) methods, as shown by several experiments on three available datasets. The source code of the proposed CRNet is available publicly at https://github.com/hilitteq/CRNet.git.

Compared to conventional video standards, the main features of H.264 standard are its high coding efficiency and its network friendliness. In spite of these outstanding merits, it is not easy to implement H.264 codec as a real-time system, due to its requirements of large memory bandwidth and intensive computation. Although the variable-block-size motion compensation using multiple reference frames is one of the key coding tools to bring about its main performance gain, its optimal use demands substantial computation for the rate-distortion calculation of all possible combinations of coding modes and estimation of the best motion vector. Many existing fast motion estimation algorithms are not suitable for H.264, which employs variable motion block sizes. We propose an adaptive motion search scheme utilizing the hierarchical block structure based on the deviation of subblock motion vectors. The proposed fast scheme adjusts the search center and search pattern according to the subblock motion-vector distribution.