The advancement of deep convolutional neural networks (CNNs) has resulted in remarkable achievements in image super-resolution methods utilizing CNNs. However, these methods have been limited by a narrow perceptual field and often require a high number of parameters and computational complexity, making them unsuitable for resource-constrained devices. Recently, the Transformer architecture has shown significant potential in image super-resolution due to its ability to perceive global features. Yet, the quadratic computational complexity of self-attention mechanisms in these Transformer-based methods leads to substantial computational and parameter overhead, limiting their practical application. To address these challenges, we introduce the Dual Branch Attention Network (DBAN), a novel Transformer model that integrates prior knowledge from traditional dictionary learning with the global feature perception capabilities of Transformers, enabling image super-resolution. Our model features a ”token dictionary” mechanism that uses auxiliary labeling to provide external prior information, enhancing cross-attention and self-attention computations while maintaining a linear relationship between computational complexity and image size. We also propose a Feature Aggregation Module (FAM) that efficiently extracts local contextual information and performs channel feature fusion, substantially enhancing the model’s performance and efficiency. By reasonably arranging the number of modules and the depth of the network, we reduce the complexity of the model. Extensive experiments have demonstrated that our DBAN achieves excellent performance.

The purpose of semi-supervised boosting strategy is to improve the classification performance of one given classifier for a large number of unlabeled data. In the semi-supervised boosting strategy, the unlabeled samples are assigned for pseudo labels according to similarities between the labeled samples and the unlabeled samples, and the unlabeled samples with high confidences of pseudo labels are selected as labeled samples at the same time. Good similarities help to assign more appropriate pseudo labels to the unlabeled samples. These selected samples with pseudo labels will be used as the labeled samples to train the new ensemble classifier. Therefore, good and distinguishable similarities learning between unlabeled samples and labeled samples has shown remarkable importance due to its promising performance for semi-supervised boosting strategy. This paper presents semi-supervised boosting framework using similarity learning based on modular sparse representation by employing a marginal regression function with probabilistic graphical structure adaptation. In this paper, distinguishable regression targets analysis, graph structure adaptation, robust modular sparse representation and semi-supervised boosting learning are seamlessly incorporated into a joint framework. This framework learns marginal regression targets from data rather than exploiting the conventional zero-one matrix that greatly hinders the freedom of regression fitness and degrades the performance of regression results to improve the interclass separation of the learned representation. Meanwhile, a regularization term based on probabilistic connection knowledge is used to construct a graph regularization with adaptive optimization, which improves the intra-class compactness of the learned representation. Additionally, modular sparse representation learning is used to improve the robustness of the learned representation. The experimental results on four datasets including face and object show that the recognition rates of the proposed method are significantly better than other state-of-the-art methods.

As recently newly techniques, two-phase sparse representation algorithms have been presented, which achieve an excellent performance in face recognition via different phase sparse representation, capturing more local structural information of samples. However, there are some defects in these algorithms:1) The Euclidean distance metric applied in these algorithms fails to capture nonlinear structural information, leading to that the performance of these algorithms is sensitive to the geometric structure of facial images. 2) To select the m nearest neighbors of the test sample is achieved directly by applying sparse representation in training samples, which ignores prior information to construct the sparse representation model. In order to solve these problems, a Weighted Two-Phase Supervised Sparse Representation based on Gaussian (GWTPSSR) algorithm is proposed on basic of existing two-phase sparse representation algorithm, in which the nonlinear local information of samples is captured by exploiting effectively the Gaussian distance metric instead of the Euclidean distance metric. Besides, GWTPSSR recreates reconstruction set from training samples in the sparse representation model for each test sample, making full use of prior information to eliminate some training samples far from the test sample. Compared with existing two-phase sparse representation algorithms, experimental results on standard face datasets show that GWTPSSR has better robustness and classification performance.