This study presents a novel character-level writer verification framework for ancient manuscripts, employing a building-block approach that integrates decision strategies across multiple token levels, including characters, words, and sentences. The proposed system utilized edge-directional and hinge features along with machine learning techniques to verify the hands that wrote the Great Isaiah Scroll. A custom dataset containing over 12,000 samples of handwritten characters from the associated scribes was used for training and testing. The framework incorporated character-specific parameter tuning, resulting in 22 separate models and demonstrated that each character has distinct features that enhance system performance. Evaluation was conducted through soft voting, comparing probability scores across different token levels, and contrasting the results with majority voting. This approach provides a detailed method for multi-scribe verification, bridging computational and paleographic methods for historical manuscript studies.
To effectively solve the problems of insufficient brightness, poor contrast, and high noise in low-light environments, this paper proposes a low-light image enhancement method based on multiscale feature fusion. The multilevel features of images are extracted by convolution kernels of different scales, and these multiscale features are fused organically by a feature fusion module. Finally, combining the processing technology of light enhancement and noise suppression, the visual effect of low-light image is significantly improved. The experimental results show that the proposed method has excellent performance in improving image brightness, contrast, and detail information, can effectively suppress noise, and has good adaptability and robustness.
This study utilizes factor extraction method and factor analysis to investigate the demand position of China Brand IP emotional preference, and build a reference system suitable for China Brand IP image. Using a semantic-emotional word classification method, the group’s emotional perception preferences for brand IP images were systematically validated. The 7-point Likert scale analysis was used to extract the semantic meanings of three public factors, namely “emotional experience factor”, “trendy factor” and “personality style factor”. These three public factors explain the emotional perception preferences of local consumers towards representative brand IP images. This study also analyses the emotional cognitive positioning of brand IP images and provides effective design references for brand IP images towards marketing strategies and optimizing design.
Based on the guidance of the national industrial by-product gypsum resource utilization policy, a set of gypsum mold box automated molding system was studied, which overcame the major industry problems of complex three-dimensional structure gypsum profile one-pouring process and the difficulty of continuous matching molding of multi-module production system and realized the stable convergence of process tasks and high-speed operation of production modules. The structural adaptation scheme for the production modules of intermittent pulping, rotary molding and mold box finishing in three-layer space was designed, which automatically adapts to complete all 11 processes of continuous molding of gypsum mold box. The actuators and control models of the production modules focus on the use of mechanical bionic technology, and are adapted to the multi-level visual human-machine interaction operating system, which realizes highly accurate and flexible operation and collaborative processing capability, and has intelligent manufacturing characteristics of workshop-level industrial robots.
Light fields (LFs) capture both angular and spatial information of light rays, providing an immersive and detailed representation of the visual world. However, the high dimensionality of LF data presents challenges for compression and transmission algorithms, which often introduce degradations that affect visual quality. To address this, we propose GCNN-LFIQA, a novel no-reference LF image quality assessment method that leverages the power of deep graph convolutional neural networks (GCNNs). The method employs a single-stream deep GCNN architecture to model the complex structural and geometric relationships within LF data, enabling accurate quality predictions. A key innovation of the proposed approach is its input preparation pipeline, which converts horizontal epipolar plane images into skeleton-based graph representations enriched with node-level features such as betweenness centrality. These graph representations serve as input to the GCNN, which predicts quality scores using a regression block. We evaluated GCNN-LFIQA on two widely used LF quality datasets, Win5-LID and LFDD, where it achieved high correlation values and outperformed other state-of-the-art methods. The proposed method demonstrates robustness, computational efficiency, and the potential to address the unique challenges of LF image quality assessment in real-world applications.
As a porous sandwich plate system, the honeycomb paperboard is widely used in cushioning packaging of products due to its energy-absorbing properties. The drop impact on a honeycomb paperboard with cell sizes of 4 mm, 6 mm, and 8 mm was simulated and experimented to investigate the paperboard’s energy absorption and deformation modes. The peak impact force of the impactor with velocity 5.47 m s−1, the dent depth of the honeycomb paperboard, and the absorbed energy were obtained. The peak forces corresponding to the core layer cell sizes of 4 mm, 6 mm, and 8 mm were 141.2 N, 108.7 N, and 97.7 N, and the absorbed energy values were 76%, 83%, and 84%, respectively. The model of honeycomb paperboard under drop impact was established. The plastic deformation and force distribution showed good agreement with experimental results, which verified the accuracy of the model. The results are expected to provide a reference for the lightweight structure and protective design of the honeycomb paperboard.
To boost the security of color image encryption algorithms and enlarge their key space, an encryption algorithm of color image based on cellular neural networks (CNNs) is proposed. The sequence produced by the 6D CNN system is segmented into two groups and combined at a specific ratio. The new chaotic sequence obtained is used as the key source for a 4D hyperchaos system. The key is selected based on the logical operation results of the plaintext pixel mean, and the final chaotic encryption sequences X and Y are obtained. Pixel scrambling, diffusion on each layer of R, G, B, and pixel value replacement encryption operations are performed on color images, which are encrypted as ciphertext images. The results of the experimental simulation demonstrate that the image encryption algorithm outlined in this paper possesses significant key space, robust sensitivity to both keys and plaintext, uniform distribution of ciphertext pixels, and a correlation coefficient near 0 among neighboring pixels. It is capable of effectively thwarting exhaustive attacks, statistical analysis attacks, and differential attacks, and produces a notable encryption impact on color images. It possesses specific utility in the realm of color image information security.
Image compression is an essential technology in image processing as it reduces video storage, which is increasingly popular. Deep learning-based image compression has made significant progress, surpassing traditional coding and decoding approaches in specific cases. Current methods employ autoencoders, typically consisting of convolutional neural networks, to map input images to lower-dimensional latent spaces for compression. However, these approaches often overlook low-frequency information, leading to sub-optimal compression performance. To address this challenge, this study proposed a novel image compression technique, Transformer and Convolutional Dual Channel Networks (TCDCN). This method extracts both edge detail and low-frequency information, achieving a balance between high and low-frequency compression. The study also utilized a variational autoencoder architecture with parallel stacked transformer and convolutional networks to create a compact representation of the input image through end-to-end training. This content-adaptive transform captured low-frequency information dynamically, leading to improved compression efficiency. Compared to the classic JPEG method, our model showed significant improvements in Bjontegaard Delta rate up to 19.12% and 18.65% on Kodak and CLIC test datasets, respectively. These improvements also surpassed the state-of-the-art solutions by notable margins of 0.47% and 0.74%, signifying a substantial enhancement in the image compression encoding efficiency. The results underscore the effectiveness of our approach in enhancing the capabilities of existing techniques, marking a significant step forward in the field of image compression.
In recent years, with the rapid development of digital printing technology, an increasing number of counterfeit products have entered the market. The anti-counterfeiting technique for QR Codes has been attracting increasing attention nowadays. There exist many image inpainting methods that can be applied in the image restoration field. Some image completion methods may restore coating QR Code images to a point where the covered digital number underneath is revealed even though the original coating QR Code has not been scratched off. In this paper, we extend the pluralistic image completion (PIC) method to scratched coating QR Code image restoration. Based on the binary characteristic of QR Codes, we present a specific type of deep learning model for scratched coating QR Code image completion. Experimental results demonstrate that the extended PIC is an effective approach to the restoration of scratched coating QR Code images.
Taking the Jurong Dongshan River as an illustrative case, we employed a Dajiang Phantom 4 RTK SE unmanned aerial vehicle (UAV) for single-lens tilt photography. Our investigation focused on examining the influence of varying flight altitudes (FA: 30 m, 60 m and 90 m) and the configuration of photo control points on two-dimentional (2D) or three-dimensional (3D) mapping accuracy of river embankments, slopes, and hydraulic structures, as well as the analysis of outcomes of 2D and 3D modeling under different FA conditions augmented with supplementary photographs. Our observations revealed that the longitude and elevation accuracy at 30 m FA were higher without photo control operations compared to those at 60 m and 90 m FA, and accuracy diminished as FA increased. Specifically, the longitude accuracy of the embankment photo control points exceeded that of the slope photo control points, whereas the elevation accuracy of the embankment photo control points was superior at FAs of 60 m and 90 m. The geographical location deviation of hydraulic structures (irrigation intake gates) in the 2D model was larger than that obtained in the 3D model. Notably, the incorporation of additional detailed photographs significantly augmented the modeling efficacy of UAV aerial survey data, especially in capturing intricate plant and slope details. It is recommended that the Phantom 4 RTK SE be used at FA of 90 m to establish a foundational channel model, along with capture of additional detailed photographs of crucial structures, slopes, etc., and obtaining geographic location information in 3D Models.