Contrastive Language–Image Pre-training (CLIP) exhibits strong zero-shot and few-shot transfer, yet its effectiveness for few-shot sketch classification remains underexplored due to a pronounced domain gap: sketches are sparse, abstract, and largely devoid of color and texture cues. We present a systematic evaluation of vanilla CLIP and recent parameter-efficient fine-tuning (PEFT) methods across multiple sketch benchmarks. To facilitate further study, we introduce Wartegg Test Classification (WTC), a new dataset derived from spontaneous psychological drawings. Finally, we propose Sketch-CLIP, a structure-aware adaptation framework that improves the efficiency–accuracy trade-off, delivering competitive gains over strong baselines. The code is available at: https://github.com/Richiexxxu/SketchCLIP .

This paper proposes a multi-task self-supervised learning framework for scene text script identification, aimed at addressing the challenges posed by diverse fonts, complex backgrounds, low resolutions, and frequent distortions in natural scenes. By leveraging unlabeled data, our approach learns robust image representations tailored for script identification. Three complementary tasks are designed: an advanced Jigsaw puzzle task to capture both local and global features, a spatial alignment-enhanced SwAV task to generalize across transformations while retaining spatial details, and a rotation prediction task to enhance spatial reasoning. Experiments on four benchmark datasets demonstrate that our method achieves state-of-the-art results, outperforming existing approaches. Ablation studies confirm the effectiveness of each module within our framework, showcasing its potential to reduce reliance on labeled data and enhance script identification in real-world applications. The code can be accessed at https://github.com/jin-or-king/Enhancing_scene_text_script_identification_through_multi-task_self-supervised_learning .

This paper is devoted to numerical algorithms based on harmonic transformations with two goals: (1) face boundary formulation by blending techniques based on the known characteristic nodes and (2) some challenging examples of face resembling. The formulation of the face boundary is imperative for face recognition, transformation, and combination. Mapping between the source and target face boundaries with constituent pixels is explored by two approaches: cubic spline interpolation and ordinary differential equation (ODE) using Hermite interpolation. The ODE approach is more flexible and suitable for handling different boundary conditions, such as the clamped and simple support conditions. The intrinsic relations between the cubic spline and ODE methods are explored for different face boundaries, and their combinations are developed. Face combination and resembling are performed by employing blending curves for generating the face boundary, and face images are converted by numerical methods for harmonic models, such as the finite difference method (FDM), the finite element method (FEM) and the finite volume method (FVM) for harmonic models, and the splitting-integrating method (SIM) for the resampling of constituent pixels. For the second goal, the age effects of facial appearance are explored to discover that different ages of face images can be produced by integrating the photos and images of the old and the young. Then, the following challenging task is targeted. Based on the photos and images of parents and their children, can we obtain an integrated image to resemble his/her current image as closely as possible? Amazing examples of face combination and resembling are reported in this paper to give a positive answer. Furthermore, an optimal combination of face images of parents and their children in the least-squares sense is introduced to greatly facilitate face resembling. Face combination and resembling may also be used for plastic surgery, finding missing children, and identifying criminals. The boundary and numerical techniques of face images in this paper can be used not only for pattern recognition but also for face morphing, morphing attack detection (MAD), and computer animation as Sora to greatly enhance further developments in AI.

Despite recent significant advancements in Handwritten Document Recognition (HDR), the efficient and accurate recognition of text against complex backgrounds, diverse handwriting styles, and varying document layouts remains a practical challenge. Moreover, this issue is seldom addressed in academic research, particularly in scenarios with minimal annotated data available. In this paper, we introduce the DocTTT framework to address these challenges. The key innovation of our approach is that it uses test-time training to adapt the model to each specific input during testing. We propose a novel Meta-Auxiliary learning approach that combines Meta-learning and self-supervised Masked Autoencoder (MAE). During testing, we adapt the visual representation parameters using a self-supervised MAE loss. During training, we learn the model parameters using a meta-learning framework, so that the model parameters are learned to adapt to a new input effectively. Experimental results show that our proposed method significantly outperforms existing state-of-the-art approaches on benchmark datasets.

Geometric image transformations are fundamental to image processing, computer vision and graphics, with critical applications to pattern recognition and facial identification. The splitting-integrating method (SIM) is well suited to the inverse transformation T−1 of digital images and patterns, but it encounters difficulties in nonlinear solutions for the forward transformation T. We propose improved techniques that entirely bypass nonlinear solutions for T, simplify numerical algorithms and reduce computational costs. Another significant advantage is the greater flexibility for general and complicated transformations T. In this paper, we apply the improved techniques to the harmonic, Poisson and blending models, which transform the original shapes of images and patterns into arbitrary target shapes. These models are, essentially, the Dirichlet boundary value problems of elliptic equations. In this paper, we choose the simple finite difference method (FDM) to seek their approximate transformations. We focus significantly on analyzing errors of image greyness. Under the improved techniques, we derive the greyness errors of images under T. We obtain the optimal convergence rates O(H2)+O(H/N2) for the piecewise bilinear interpolations (μ=1) and smooth images, where H(≪1) denotes the mesh resolution of an optical scanner, and N is the division number of a pixel split into N2 sub-pixels. Beyond smooth images, we address practical challenges posed by discontinuous images. We also derive the error bounds O(Hβ)+O(Hβ/N2), β∈(0,1) as μ=1. For piecewise continuous images with interior and exterior greyness jumps, we have O(H)+O(H/N2). Compared with the error analysis in our previous study, where the image greyness is often assumed to be smooth enough, this error analysis is significant for geometric image transformations. Hence, the improved algorithms supported by rigorous error analysis of image greyness may enhance their wide applications in pattern recognition, facial identification and artificial intelligence (AI).

This study examines the integration of graphology and machine learning, including deep learning, to analyze personality traits from handwriting. The research builds on the premise that neuromuscular movements in handwriting reflect personality characteristics. A dataset of 1,108 handwriting images, sourced from the Centre for Pattern Recognition and Machine Intelligence (CENPARMI) and a graphology expert, was analyzed using machine learning algorithms such as k-Nearest Neighbor (k-NN), Random Forest, Logistic Regression, and deep learning techniques via transfer learning. Data balancing was addressed using the Synthetic Minority Over-Sampling Technique (SMOTE), and ensemble methods like Majority Voting and Stacking improved classification performance. Results showed over 90

Progressive loss of motor, cognitive, or both capacities is a hallmark of neurodegenerative disorders including Parkinson's and Alzheimer's. Although the precise causes of neurodegenerative disorders are frequently unknown, a breakdown in brain and spinal cord function is the scientific interpretation of these conditions, which can be caused by genetic, environmental, and lifestyle factors. Since there is currently no cure for these disorders, treatment usually concentrates on managing symptoms in order to enhance patients' quality of life and delay the course of the disease. Given that serious and irreparable harm may have already been done once symptoms appear, an early diagnosis is essential to begin medical care right away. Handwriting, which depends on a mix of kinesthetic and motor-perceptual skills, is commonly recognized as one of the first skills impacted by cognitive problems. The development of several handwriting protocols that describe the exact writing or drawing tests to be conducted is the first step toward the significant improvements that have been made in the field. However, it is crucial to emphasize that there is no universal agreement on the number and type of tasks that should be used. Furthermore, there aren't many standardized databases that compile this type of data, which usually only pertains to a small number of people. This aspect adds another layer of complexity in the realm of machine learning techniques, which usually demand substantial volumes of data. Moreover, there is a lack of consensus regarding the specific features researchers should prioritize. Indeed, the challenge of identifying effective features that enable the system to differentiate between regular age-related handwriting changes and those induced by neurodegenerative disorders remains unresolved. This paper will investigate existing studies focusing on detecting Alzheimer's disease using handwriting analysis. Our examination will encompass the databases employed, the features extracted, the methodologies applied, and the ultimate discoveries and conclusions drawn from these studies.

Art style classification plays a crucial role in computational aesthetics. Traditional deep learning-based methods for art style classification typically require a large number of labeled images, which are scarce in the art domain. To address this challenge, we propose a self-supervised learning method specifically tailored for art style classification. Our method effectively learns image style features using unlabeled images. Specifically, we introduce a novel self-supervised learning approach based on the popular contrastive learning framework, incorporating a unique dual-teacher knowledge distillation technique. The two teacher networks provide complementary guidance to the student network. Each teacher network focuses on extracting distinct features, offering diverse perspectives. This collaborative guidance enables the student network to learn detailed and robust representations of art style attributes. Furthermore, recognizing the Gram matrix's capability to capture image style through feature correlations, we explicitly integrate it into our self-supervised learning framework. We propose a relation alignment loss to train the network, leveraging image relationships. This loss function has shown promising results compared to the commonly used InfoNCE loss. To validate our proposed method, we conducted extensive experiments on three publicly available datasets: WikiArt, Pandora18k, and Flickr. The experimental results demonstrate the superiority of our method, significantly outperforming state-of-theart self-supervised learning methods. Additionally, when compared with supervised methods, our approach shows competitive results, notably surpassing supervised learning methods on the Flickr dataset. Ablation experiments further verify the efficacy of each component of our proposed network. The code is publicly available at: https://github.com/lm-oc/dual_signal_gram_matrix.

Developing a highly accurate automatic license plate recognition system (ALPR) is challenging due to environmental factors such as lighting, rain, and dust. Additional difficulties include high vehicle speeds, varying camera angles, and low-quality or low-resolution images. ALPR is vital in traffic control, parking, vehicle tracking, toll collection, and law enforcement applications. This paper proposes a deep learning strategy using YOLOv8 for license plate detection and recognition tasks. This method seeks to enhance the performance of the model using datasets from Ontario, Quebec, California, and New York State. It achieved an impressive recall rate of 94

In the field of handwriting analysis, deep neural networks seemed to perform not as well as the classic machine-learning methods (such as the support vector machine). Thus, we started the inspection from the source, that is raw data, the input of deep neural networks. One is random splits, it has a positive effect on test accuracy and loss, solving the small sample problem by creating more samples to make up for the lack of training samples. The other is preprocessing. We carried out experiments with and without preprocessing, and the results show that preprocessing operation is necessary especially binarization operations. Furthermore, the order of preprocessing operation also affects the final prediction result. Finally, we compared with state-of-the-art methods, and obtained competitive experimental results, discovering the power of micro-inspection in handwriting analysis.

This paper examines the impact of input data structure on deep learning models for handwriting analysis, focusing on the concept of “granularity.” Granularity represents the degree of segmentation or the perspective through which models process input data. Our findings show that splitting operations not only alleviate the issue of small sample sizes but also help identify the ideal zoom scale for model training. The key to achieving optimal granularity lies in balancing detailed local information with broader global patterns, rather than simply increasing the number of data splits. When selected effectively, granularity can reduce the dependency on additional training samples. Our experimental results indicate that the optimal granularities are 3 × 8 (horizontal strips), 8 × 6 (vertical strips), and 9 × 12 (square segments). Based on these findings, we recommend that handwriting analysis training samples consist of 3 to 4 text lines, each containing more than 2 words per line, with training sample sizes ranging from 77 to 1506.

The intersection of Artificial Intelligence (AI) and the Internet of Things (IoT) has propelled the agricultural industry into a new era of efficiency and sustainability. Among the diverse applications of AI and IoT in agriculture, Smart Greenhouses (SGHs) are particularly notable for their transformative potential in revolutionizing crop cultivation practices. Moreover, the adoption of SGH technologies has significant implications for agricultural sustainability and environmental conservation. By minimizing resource waste and reducing reliance on chemical inputs, SGHs mitigate the environmental impact of traditional farming practices. The aim of this comprehensive survey is evaluating the state-of-the-art literature on SGH development using AI and sensory data. In addition, this survey is one of the first to bridge the gap between academic research and industrial applications of AI-powered SGHs, offering a holistic view of the field’s progress and future prospects. This work also critically examines the technical level of the surveyed works and their alignment with the current AI trends. This comprehensive survey follows a well-defined review protocol and inclusion criteria. A total of 88 studies, industrial projects, related datasets from different research sources, namely, IEEE, SpringerLink and Science Direct were included in the review. The survey critically assesses both academic and industrial SGH projects, identifying key research gaps and the lag in adopting recent AI innovations.

In the field of historical manuscript research, scholars frequently encounter novel symbols in ancient texts, investing considerable effort in their identification and documentation. Although existing object detection methods achieve impressive performance on known categories, they struggle to recognize novel symbols without retraining. To address this limitation, we propose a Visually Guided Text Spotting (VGTS) approach that accurately spots novel characters using just one annotated support sample. The core of VGTS is a spatial alignment module consisting of a Dual Spatial Attention (DSA) block and a Geometric Matching (GM) block. The DSA block aims to identify, focus on, and learn discriminative spatial regions in the support and query images, mimicking the human visual spotting process. It first refines the support image by analyzing inter- channel relationships to identify critical areas, and then refines the query image by focusing on informative key points. The GM block, on the other hand, establishes the spatial correspondence between the two images, enabling accurate localization of the target character in the query image. To tackle the example imbalance problem in low-resource spotting tasks, we develop a novel torus loss function that enhances the discriminative power of the embedding space for distance metric learning. To further validate our approach, we introduce anew dataset featuring ancient Dongba hieroglyphics (DBH) associated with the Naxi minority of China. Extensive experiments on the DBH dataset and other public datasets, including Egyptian Hieroglyph (EGY), Historical Arabic Documents (HAD), Tripitaka Koreana in Han (TKH), and Notary Charters (NC), show that VGTS consistently surpasses state-of-the-art methods. The proposed framework exhibits great potential for application in historical manuscript text spotting, enabling scholars to efficiently identify and document novel symbols with minimal annotation effort.

Counterfeit coins pose a significant challenge in various real-world applications, from vending machines to currency exchange systems, making their reliable detection a matter of utmost importance. This research presents a novel framework designed to tackle this issue by harnessing the power of image-mining techniques. Our proposed framework is developed in two modules. In the first module, a method to detect the region of interest (ROIs) is applied that focuses on blob detection. In the second module, image mining is applied to find image patterns present in coin images using fuzzy association rules mining. The enhancement lies in utilizing Particle Swarm Optimization (PSO) within the image mining module. PSO refines the threshold parameters, thereby improving the efficiency of the fuzzy association rules mining process. This integration allows for the automatic determination of optimal values, contributing to the overall robustness of the counterfeit coin detection system. Comprising two modules, this framework offers a unique advantage as a compress, serving as a knowledge attainment tool. By harnessing the full power of fuzzy association rule mining, this paper introduces pruning methods to reduce redundant and insignificant rules. Moreover, we propose a novel algorithm for feature selection and a pruned-based fuzzy associative classifier to establish a robust counterfeit coin detection system. Comparative analysis with other methods using the same dataset showcases the superiority of our framework, exhibiting lower feature dimensions, smoother boundaries, and maintaining satisfactory accuracy. The generality of this study's problem formulation offers a common framework for addressing similar challenges across various domains.

Humans often require only a few visual archetypes to spot novel objects. Based on this observation, we present a strategy rooted in ``spotting the unseen" by establishing dense correspondences between potential query image regions and a visual archetype, and we propose the Consensus Network (CoNet). Our method leverages relational patterns intra and inter images via Auto-Correlation Representation (ACR) and Mutual-Correlation Representation (MCR). Within each image, the ACR module is capable of encoding both local self-similarity and global context simultaneously. Between the query and support images, the MCR module computes the cross-correlation across two image representations and introduces a reciprocal consistency constraint, which can incorporate to exclude outliers and enhance model robustness. To overcome the challenges of low-resource training data, particularly in one-shot learning scenarios, we incorporate an adaptive margin strategy to better handle diverse instances. The experimental results indicate the effectiveness of the proposed method across diverse domains such as object detection in natural scenes, and text spotting in both historical manuscripts and natural scenes, which demonstrates its sparkling generalization ability. Our code is available at: https://github.com/infinite-hwb/conet.

Reticulocyte count is a routine blood test that can be an essential source of knowledge for medical doctors to diagnose and assess patients’ health condition. In fact, the automation of this blood test will reduce cost and time, in addition to protecting laboratorians’ lives, especially during pandemics and outbreaks. However, human reticulocyte data scarcity is a main challenge that slows the pace of the test automation. In this paper, a novel method that assesses the feasibility of using animal reticulocyte cells as a solution to compensate for the scarcity of human reticulocyte data is investigated. The integration of animal cells will be implemented by utilizing a data-centric artificial intelligence approach, in addition to employing multiple deep classifiers that utilize transfer learning in different experimental setups in a procedure that mimics the protocol followed in experimental medical labs. Moreover, to evaluate the effectiveness of the proposed method, three evaluation criteria have been proposed, namely, the pretraining boost, the dataset similarity boost, and the dataset size boost measures. All the experiments of this work were conducted on a public human reticulocyte dataset and the best performing model achieved 98.9%, 98.9%, 98.6% average accuracy, average macro precision, and average macro F-score respectively. Moreover, the results showed that using animals medical data holds a promising solution for human medical data scarcity, as utilizing weights that were pretrained on a medium size feline reticulocyte dataset outperformed the model that utilized weights that were pretrained on the large scale ImageNet dataset.

Handwriting analysis has traditionally been used to infer personality traits from the stylistic features of writing. With advances in machine learning, the accuracy and applicability of these analyses have significantly improved. This paper presents a new multi-label classification approach to classify personality traits, such as Extraversion and Conscientiousness, into Low, Average, and High categories. This approach uses Binary Cross-Entropy with Logits Loss and Focal Loss to handle multi-label classification and class imbalance. Image segmentation techniques are also employed to enhance the handling of limited handwriting samples. The paper evaluates the complexity and performance of ResNet-50 and ResNet-101 architectures in recognizing complex handwriting patterns using three optimizers: SGD with momentum, Adam, and AdaBelief. The results demonstrate the efficacy of our proposed method, improving overall accuracy from 67.09

License plate detection (LPD) is essential for traffic management, vehicle tracking, and law enforcement but faces challenges like variable lighting and diverse font types, impacting accuracy. Traditionally reliant on image processing and machine learning, the field is now shifting towards deep learning for its robust performance in various conditions. Current methods, however, often require tailoring to specific regional datasets. This paper proposes a dual deep learning strategy using a Faster R-CNN for detection and a CNN-RNN model with Connectionist Temporal Classification (CTC) loss and a MobileNet V3 backbone for recognition. This approach aims to improve model performance using datasets from Ontario, Quebec, California, and New York State, achieving a recall rate of 92

Script identification is an essential preliminary step in multilingual OCR systems. This paper focuses primarily on tackling the challenging problem of script identification in scene text images, which are usually characterized by low image quality, diverse text styles, and complex backgrounds. Furthermore, script identification becomes a fine-grained classification problem when some scripts share common characters. To address this issue, we propose a novel end-to-end CNN comprising two streams for extracting distinct types of features, namely, visual features and spatial features. In the visual stream, we introduce an enhanced Squeeze-and-Excitation (SE) channel attention mechanism to emphasize valuable features and suppress irrelevant ones. The enhanced SE is composed of squeeze and excitation steps. The squeeze step employs adaptive average pooling for information aggregation. Two 1x1 convolutional layers are used to derive channel weights in the excitation step. In the spatial stream, we perform efficient analysis of the spatial dependencies within the text lines based on LSTM. Finally, we propose an adaptive fusion approach that combines probability vectors from the two streams. Instead of being fixed, the weight assigned to each probability vector is learned during network training. To validate our proposed method, we conduct extensive tests on four publicly available datasets, viz. MLe2e, RRC-MLT2017, SIW-13, and CVSI-2015. Our proposed method achieves accuracies of 97.66% , 90.24% , 96.66% , and 98.44% on these four datasets, respectively, which compare favorably with state-of-the-art methods. The two streams have demonstrated complementarity. Moreover, ablation experiments have been conducted to verify the effectiveness of each component in the proposed method.