Recent Blind Image Super-Resolution (BSR) methods have shown proficiency in general images. However, we find that the efficacy of recent methods obviously diminishes when employed on image data with blur, while images with intentional blur constitute a substantial proportion of general data. To further investigate and address this issue, we developed a new super-resolution dataset specifically tailored for blur images, named the Real-world Blur-kept Super-Resolution (ReBlurSR) dataset, which consists of nearly 3000 defocus and motion blur image samples with diverse blur sizes and varying blur intensities. Furthermore, we propose a new BSR framework for blur images called Perceptual-Blur-adaptive Super-Resolution (PBaSR), which comprises two main modules: the Cross Disentanglement Module (CDM) and the Cross Fusion Module (CFM). The CDM utilizes a dual-branch parallelism to isolate conflicting blur and general data during optimization. The CFM fuses the well-optimized prior from these distinct domains cost-effectively and efficiently based on model interpolation. By integrating these two modules, PBaSR achieves commendable performance on both general and blur data without any additional inference and deployment cost and is generalizable across multiple model architectures. Rich experiments show that PBaSR achieves state-of-the-art (SOTA) performance across various quantitative metrics without incurring extra inference costs. Within the commonly adopted Learned Perceptual Image Patch Similarity (LPIPS) setting, PBaSR achieves an improvement range of approximately 0.02-0.09 with diverse anchor methods and blur types, across both the ReBlurSR and multiple widely used general BSR benchmarks.

Text image super-resolution is a unique and important task to enhance readability of text images to humans. It is widely used as pre-processing in scene text recognition. However, due to the complex degradation in natural scenes, recovering high-resolution texts from the low-resolution inputs is ambiguous and challenging. Existing methods mainly leverage deep neural networks trained with pixel-wise losses designed for natural image reconstruction, which ignore the unique character characteristics of texts. A few works proposed content-based losses. However, they only focus on text recognizers' accuracy, while the reconstructed images may still be ambiguous to humans. Further, they often have weak generalizability to handle cross languages. To this end, we present TATSR, a Text-Aware Text Super-Resolution framework, which effectively learns the unique text characteristics using Criss-Cross Transformer Blocks (CCTBs) and a novel Content Perceptual (CP) Loss. The CCTB extracts vertical and horizontal content information from text images by two orthogonal transformers, respectively. The CP Loss supervises the text reconstruction with content semantics by multi-scale text recognition features, which effectively incorporates content awareness into the framework. Extensive experiments on various language datasets demonstrate that TATSR outperforms state-of-the-art methods in terms of both recognition accuracy and human perception.

Blind super-resolution (BSR) methods based on high-resolution (HR) reconstruction codebooks have achieved promising results in recent years. However, we find that a codebook based on HR reconstruction may not effectively capture the complex correlations between low-resolution (LR) and HR images. In detail, multiple HR images may produce similar LR versions due to complex blind degradations, causing the HR-dependent only codebooks having limited texture diversity when faced with confusing LR inputs. To alleviate this problem, we propose the Rich Texture-aware Codebook-based Network (RTCNet), which consists of the Degradation-robust Texture Prior Module (DTPM) and the Patch-aware Texture Prior Module (PTPM). DTPM effectively mines the cross-resolution correlation of textures between LR and HR images by exploiting the cross-resolution correspondence of textures. PTPM uses patch-wise semantic pre-training to correct the misperception of texture similarity in the high-level semantic regularization. By taking advantage of this, RTCNet effectively gets rid of the misalignment of confusing textures between HR and LR in the BSR scenarios. Experiments show that RTCNet outperforms state-of-the-art methods on various benchmarks by up to 0.16 ~ 0.46dB.

Molecular circuits and devices with temporal signal processing capability are of great significance for the analysis of complex biological processes. Mapping temporal inputs to binary messages is a process of history-dependent signal responses, which can help understand the signal-processing behavior of organisms. Here, we propose a DNA temporal logic circuit based on DNA strand displacement reactions, which can map temporally ordered inputs to corresponding binary message outputs. The presence or absence of the output signal is determined by the type of substrate reaction with the input so that different orders of inputs correspond to different binary outputs. We demonstrate that a circuit can be generalized to more complex temporal logic circuits by increasing or decreasing the number of substrates or inputs. We also show that our circuit had excellent responsiveness to temporally ordered inputs, flexibility, and expansibility in the case of symmetrically encrypted communications. We envision that our scheme can provide some new ideas for future molecular encryption, information processing, and neural networks.

The construction of DNA reaction networks with complex functions using various methods has been an important research topic in recent years. Whether the DNA reaction network can perform complex tasks and be recycled directly affects the performance of the reaction network. Therefore, it is very important to design and implement a DNA reaction network capable of multiple tasks and reversible regulation. In this paper, the hairpin allosteric method was used to complete the assembly task of different functional nucleic acids. In addition, information conversion of the network was realized. In this network, multiple hairpins were assembled into nucleic acid structures with different functions to achieve different output information through the cyclic use of trigger strands. A method of single-input dual-output information conversion was proposed. Finally, the network with signal amplification and reversible regulation was constructed. In this study, the reversible regulation of different functional nucleic acids in the same network was realized, which shows the potential of this network in terms of programmability and provides new ideas for constructing complex and multifunctional DNA reaction networks.

The construction of logic circuits is critical to DNA computing. Simple and effective scalability methods have been the focus of attention in various fields related to constructing logic circuits. We propose a double-stranded separation (DSS) strategy to facilitate the construction of complex circuits. The strategy combines toehold-mediated strand displacement with exonuclease III (Exo III), which is a multifunctional nuclease. Exo III can quickly recognize an apurinic/apyrimidinic (AP) site. DNA oligos with an AP site can generate an output signal by the strand displacement reaction. However, in contrast to traditional strand displacement reactions, the double-stranded waste from the strand displacement can be further hydrolysed by the endonuclease function of Exo III, thus generating an additional output signal. The DSS strategy allows for the effective scalability of molecular logic circuits, enabling multiple logic computing capabilities simultaneously. In addition, we succeeded in constructing a logic circuit with dual logic functions that provides foundations for more complex circuits in the future and has a broad scope for development in logic computing, biosensing, and nanomachines.

Blind super-resolution (BSR) methods based on high-resolution (HR) reconstruction codebooks have achieved promising results in recent years. However, we find that a codebook based on HR reconstruction may not effectively capture the complex correlations between low-resolution (LR) and HR images. In detail, multiple HR images may produce similar LR versions due to complex blind degradations, causing the HR-dependent only codebooks having limited texture diversity when faced with confusing LR inputs. To alleviate this problem, we propose the Rich Texture-aware Codebook-based Network (RTCNet), which consists of the Degradation-robust Texture Prior Module (DTPM) and the Patch-aware Texture Prior Module (PTPM). DTPM effectively mines the cross-resolution correlation of textures between LR and HR images by exploiting the cross-resolution correspondence of textures. PTPM uses patch-wise semantic pre-training to correct the misperception of texture similarity in the high-level semantic regularization. By taking advantage of this, RTCNet effectively gets rid of the misalignment of confusing textures between HR and LR in the BSR scenarios. Experiments show that RTCNet outperforms state-of-the-art methods on various benchmarks by up to 0.16 ~ 0.46dB.

With the advent of the era of big data, personalized recommendation is becoming more and more important. However, there is a problem of data sparsity in recommender system. The sparseness of data leads to the inaccuracy of recommendation. With the rapid development of social network, researchers begin to add social relationship into recommender system. Trust relationship is a very important kind of relationships. In this paper, we bring up a new approach for rating prediction by using trust computation. First, we describe a method for trust computation. Then, we give the strategy of prediction by using trust and user's rating preference. Finally, we compare our model with four related methods and conclude that our model can not only improve the prediction effect but also recommend more kinds of items accurately.