Image Contour Research Articles

Detection and quantification of small and low-uptake lesions for differentiated thyroid carcinoma using non-time-of-flight iodine-124 PET/MRI.

124-iodine (124I) is used for positron emission tomography (PET) diagnostics and therapy planning in patients with differentiated thyroid cancer (DTC). Small lesion sizes(<10mm) and low 124I uptake are challenging conditions for the detection of DTC lymph node lesions. The aim of this study was to systematically investigate the lesion detectability and quantification performance under clinically challenging imaging conditions using non-time-of-flight (TOF) PET/magnetic resonance imaging (MRI) in the clinical context of radionuclide therapy planning of DTC patients. PET/MR measurements were performed on the Siemens Biograph mMR using a small lesion NEMA-like phantom (six glass spheres, diameters 3.7-9.7mm). 60 min list-mode data were acquired for nine activity concentrations (AC) ranging from 25kBq/mL to 0.25kBq/mL using a sphere-to-background ratio of 20:1. PET list-mode data were divided into five timeframes (60, 30, 16, 8, and 4min) and reconstructed using either ordered-subsets expectation maximization (OSEM) or OSEM+ point spread function (PSF) algorithm. For all reconstructions, the smallest detectable sphere size was investigated in a human observer study. Partial volume effect (PVE) corrected PET images (contour and oversize-based approach) were analyzed considering a±30% deviation range between imaged and true AC as acceptable. Clinical data of eight DTC patients with small lymph node lesions were evaluated to assess agreement between the PVE correction approaches. Longer PET acquisition times, higher ACs, and PSF reconstructions resulted in improved PET image quality and overall improved lesion detectability. The smallest 3.7mm sphere was only visible under the best imaging conditions. Using a typical clinical 124I whole-body PET/MRI protocol with an acquisition time of 8 min using OSEM reconstructions, all lesions of ≥6.5mm in diameter could be detected and the quantification provided reliable results approximately above 5.0kBq/mL. An accurate quantification of ACs in the 4.8mm sphere was not feasible in this study. In the clinical evaluation of 10 lesions, a good agreement between oversize- and contour-based PVE corrections was observed(<15% deviation). The results showed that a reliable quantification of 124I uptake with PET/MRI is feasible and, therefore, could be used to perform radioiodine pre-therapy lesion dosimetry and individualized therapy planning in DTC patients.

Development of a novel automated algorithm for patient dosimetry in computed tomography: a step towards the facilitation of size-specific dose estimates and organs dosimetry estimations in a busy clinical workflow

Abstract Accurate dosimetry in computed tomography (CT) is essential for patient safety and effective radiation management. This study presents the development of an automated algorithm designed to enhance patient dosimetry by facilitating size-specific dose estimates (SSDE) and organ dose estimations. Utilizing a Python-based script, the proposed method integrates advanced image preprocessing, contour detection, and mathematical calculations to quantify key metrics from CT images. This automated approach addresses the limitations of manual measurement techniques. A retrospective analysis was conducted on CT axial images from examinations acquired with an 80-detector scanner. The algorithm processes DICOM images, converts pixel values to Hounsfield Units, applies Gaussian smoothing, windowing, and thresholding, followed by morphological operations to refine segmentation. It measures the water equivalent diameter (Dw) and estimates both region SSDE and organ doses, incorporating tissue attenuation. Validation was performed using an adult anthropomorphic ATOM phantom, with organ doses measured by optically stimulated luminescence dosimeters. The results demonstrated the algorithm’s potential in estimating SSDE and organ doses. Validation of the automated method revealed strong correlations for Dw and SSDE between the proposed method and manual measurements of five expert reviewers ranging from 0.86 to 0.99 for determination coefficient. Comparative analysis of organ doses showed close agreement between results from experimental setup against the proposed algorithm. The automated algorithm estimated brain dose with a mean of 21.8 mGy, while measurements from the ATOM phantom and CT Expo indicated 19.74 mGy and 23.05 mGy, respectively. For lung doses, the automated algorithm estimated 12.5 mGy compared to 11.0 mGy from the ATOM phantom and 13.1 mGy from CT Expo. Liver doses were measured at 12.7 mGy by the automated method, versus 12.1 mGy from the ATOM phantom and 11.1 mGy from CT Expo. This study shows the potential of automated image analysis techniques in enhancing dosimetry accuracy in CT examinations.

Development of a seat support mechanism for guiding users to an appropriate posture in a seated-style echocardiography robot

Abstract This paper proposes a seat support mechanism to solve the problem of misalignment between the chest examination and probe scanning areas when the patient is bent in a seated-style echocardiography robot. To guide the patient to an appropriate body position where their chest is within the examination range of the chest examination unit while minimizing the physical load of the patient, the posture of the patient must satisfy the three following conditions. (i) The breech must be in contact with the seat surface, (ii) the legs must be vertical to the floor, and (iii) the chest and mechanism must be parallel while the probe scanning and chest examination ranges must match. The human body was modeled to derive a posture that satisfies the aforementioned conditions for the height of each individual, and a seat support mechanism with four degrees of freedom was installed to guide the user to the derived posture. By installing this mechanism, the body load of the left biceps brachii, right biceps brachii, left latissimus dorsi, and right latissimus dorsi was reduced to 64.7%, 52.7%, 86.4%, and 80.2%, respectively. The sharpness of the image contours was improved to 103.8%.

Computer vision-based intelligent detection method for the residual capability of energy dissipators in flexible protection systems

A residual capability intelligent detection method based on computer vision is proposed to address the issues of low efficiency, poor accuracy, and high danger in manual measurement of energy dissipators in flexible protection systems. The proposed method first establishes a binary semantic segmentation dataset for energy dissipators and trains a salient object detection deep neural network to segment the energy dissipator binary map; Then, it uses morphological image processing and contour detection to calculate the residual capability automatically. U2-Net, U2-Netp, and BASENet were trained and compared by a dataset with 500 ring-type energy dissipator images. The proposed method was validated through a quasi-static tensile test and a full-scale impact test. Compared with the most accurate integration calculation method, the error of the proposed method does not exceed 3 %, and the efficiency is improved by about 25 times compared to the most commonly used manual detection method.

Deep learning-powered efficient characterization and quantification of microplastics

Characterizing and quantifying microplastics (MPs) are time-consuming and labor-intensive tasks traditionally. This paper presents an artificial intelligence (AI) framework aiming to automate these tasks by integrating computer vision and deep learning techniques. The approach leverages Fourier Transform Infrared (FTIR) spectra and visual images. Primary novelties of this research involve the development of: (1) an AI framework integrating efforts of data processing, analytics, visualization, and human-computer interaction; (2) a method for transforming FTIR data into contour images; (3) data augmentation strategies for resolving data scarcity and imbalance issues; (4) deep learning models for identifying MPs; (5) computer vision algorithms for quantifying MPs; and (6) an engineer-friendly graphic user interface (GUI) for enhancing data accessibility. The AI framework has been applied to polyethylene, polypropylene, polystyrene, polyamide, ethylene-vinyl acetate, and cellulose acetate. Results confirmed the efficacy of the framework, exhibiting high accuracy scores in classification (98 %), segmentation (99 %), and quantification (96 %) tasks. This research advances the capability of automatic assessment of MPs.

Salient object detection with bayesian inference based on radar and camera fusion used in UAV obstacle avoidance

Most existing salient object detection methods are sensitive to background noise and rely on prior information in UAV obstacle avoidance applications despite detection methods witnessing rapid progress. In this paper, we propose an efficient framework for salient object detection based on radar-camera fusion and iterative Bayesian optimization. A rough salient object (RSO) image is first built through radar and camera calibration. Next, the RSO image is used to calculate the edge response based on the receptive field mechanism of the primary visual cortex to construct the contour image. Finally, the above two images and the 2D Gaussian distribution are jointly integrated into an iterative Bayesian optimization scheme to get the final salient object image. Different from typical detection methods, this method suppresses background noise by filtering out irrelevant pixels using fusion information. The Bayesian inference framework’s detection performance is improved by precise spatial prior, consisting of optimized contour and RSO images. Experimental results indicate that the presented algorithm performs well against state-of-the-art salient object detection methods on several reference datasets in different evaluation metrics.

Lung cancer classification model using convolutional neural network with feature ranking process

Abstract Lung cancer is the leading cause of cancer-related deaths worldwide, highlighting the importance of early detection to improve patient outcomes. The goal of this study is to create a computer-aided diagnosis (CAD) system that detects and classifies lung cancer based on medical images using a Convolutional Neural Network (CNN) and feature extraction techniques. The study employs the LIDC-IDRI dataset, a comprehensive collection of lung cancer-related medical images, to achieve this goal. Contrast Limited Adaptive Histogram Equalization (CLAHE) is used to improve contrast and detail while reducing noise. The Sobel operator is used to refine the segmentation process by enhancing edges and contours in binary images. Morphological operations such as dilation and filling are used to refine the segmented regions even further. Feature extraction is used to extract statistical data and texture characteristics from segmented regions. Mean and variance calculations reveal information about brightness and variability within regions, whereas co-occurrence matrices and Gray-Level Co-occurrence Matrix (GLCM) properties quantify texture features. The correlation between different regions is also evaluated to assess their relationships. Using the pre-processed and ranked features as inputs, a CNN model with five hidden layers is trained. To classify the segmented regions into cancerous and non-cancerous classes, the model learns patterns and relationships in the data. A confusion matrix is used to assess the accuracy, specificity, and sensitivity of the model's predictions, with an emphasis on correctly identifying lung cancer-affected regions. The results show promising results, with the proposed CAD system identifying lung cancer-affected regions with an accuracy of 99.4375%. The system also outperforms other existing methods with a specificity of 99.12% and a sensitivity of 99.26%. These findings highlight the developed system's potential as a valuable tool for early lung cancer detection, assisting doctors in making accurate diagnoses and improving patient outcomes.

Concrete Crack Detection and Segregation: A Feature Fusion, Crack Isolation, and Explainable AI-Based Approach.

Scientific knowledge of image-based crack detection methods is limited in understanding their performance across diverse crack sizes, types, and environmental conditions. Builders and engineers often face difficulties with image resolution, detecting fine cracks, and differentiating between structural and non-structural issues. Enhanced algorithms and analysis techniques are needed for more accurate assessments. Hence, this research aims to generate an intelligent scheme that can recognize the presence of cracks and visualize the percentage of cracks from an image along with an explanation. The proposed method fuses features from concrete surface images through a ResNet-50 convolutional neural network (CNN) and curvelet transform handcrafted (HC) method, optimized by linear discriminant analysis (LDA), and the eXtreme gradient boosting (XGB) classifier then uses these features to recognize cracks. This study evaluates several CNN models, including VGG-16, VGG-19, Inception-V3, and ResNet-50, and various HC techniques, such as wavelet transform, counterlet transform, and curvelet transform for feature extraction. Principal component analysis (PCA) and LDA are assessed for feature optimization. For classification, XGB, random forest (RF), adaptive boosting (AdaBoost), and category boosting (CatBoost) are tested. To isolate and quantify the crack region, this research combines image thresholding, morphological operations, and contour detection with the convex hulls method and forms a novel algorithm. Two explainable AI (XAI) tools, local interpretable model-agnostic explanations (LIMEs) and gradient-weighted class activation mapping++ (Grad-CAM++) are integrated with the proposed method to enhance result clarity. This research introduces a novel feature fusion approach that enhances crack detection accuracy and interpretability. The method demonstrates superior performance by achieving 99.93% and 99.69% accuracy on two existing datasets, outperforming state-of-the-art methods. Additionally, the development of an algorithm for isolating and quantifying crack regions represents a significant advancement in image processing for structural analysis. The proposed approach provides a robust and reliable tool for real-time crack detection and assessment in concrete structures, facilitating timely maintenance and improving structural safety. By offering detailed explanations of the model's decisions, the research addresses the critical need for transparency in AI applications, thus increasing trust and adoption in engineering practice.

Digital bio-dosimeter for detection of deformed biological cells irradiated by ionizing radiations

In this paper, a digital bio-dosimeter (DBD) with a square window size of 7 × 7 pixels is proposed for the quantitative detection of the deformed biological cells irradiated by X-rays with different energies and γ-rays from 60Co. The basic idea of the DBD is to compute the energy of the bio-dosimeter window size by summing up the entire gray levels. Peripheral blood films (PBFs) from male albino rats irradiated by ionizing radiations are prepared and photographed using an Automatic Image Contour Analysis system with high magnification to produce a 2D image. The surf function in MATLAB is used to convert the 2D image to a 3D shaded surface to resample the date more finely. The proposed DBD shows that the computed energies on the boundaries of the normal RBCs are nearly the same, while the computed energies on the boundaries of the deformed cells are significantly larger.

Optimizing Corn Tar Spot Measurement: A Deep Learning Approach Using Red-Green-Blue Imaging and Stromata Contour Detection Algorithm for Leaf-Level Disease Severity Analysis.

Visual detection of stromata (brown-black, elevated fungal fruiting bodies) is a primary method for quantifying tar spot early in the season, as these structures are definitive signs of the disease and essential for effective disease monitoring and management. Here, we present Stromata Contour Detection Algorithm version 2 (SCDA v2), which addresses the limitations of the previously developed SCDA version 1 (SCDA v1) without the need for empirical search of the optimal Decision Making Input Parameters (DMIPs), while achieving higher and consistent accuracy in tar spot stromata detection. SCDA v2 operates in two components: (i) SCDA v1 producing tar-spot-like region proposals for a given input corn leaf Red-Green-Blue (RGB) image, and (ii) a pre-trained Convolutional Neural Network (CNN) classifier identifying true tar spot stromata from the region proposals. To demonstrate the enhanced performance of the SCDA v2, we utilized datasets of RGB images of corn leaves from field (low, middle, and upper canopies) and glasshouse conditions under variable environments, exhibiting different tar spot severities at various corn developmental stages. Various accuracy analyses (F1-score, linear regression, and Lin's concordance correlation), showed that SCDA v2 had a greater agreement with the reference data (human visual annotation) than SCDA v1. SCDA v2 achievd 73.7% mean Dice values (overall accuracy), compared to 30.8% for SCDA v1. The enhanced F1-score primarily resulted from eliminating overestimation cases using the CNN classifier. Our findings indicate the promising potential of SCDA v2 for glasshouse and field-scale applications, including tar spot phenotyping and surveillance projects.

Nakagami-fuzzy imaging for grading brain tumors by analyzing fractal complexity

Gliomas are the brain tumors in glial cells, which are categorized into four numerical grades, I-II-III-IV, to quantize the aggressiveness and severity of the tumors; while divided into two major groups, high-grade (HG) and low-grade (LG), in general. Among many differences between these groups, one of the most distinct and characteristic features could be seen in the shape of the tumor boundaries by magnetic resonance imaging (MRI). Due to aggressive nature of the HG tumors in proliferation phase, the boundaries of HG tumors become more shape-wise complex compared to the LG tumors, which could be differentiated by analyzing the fractal complexity of the cell membranes. However, the complexity cannot be either manually calculated or estimated by eye inspection without a reference point with one single image or sometimes even with an image set. Therefore, we present an automated glioma grading framework to provide an insight on the grades with a novel contouring and fractal dimension analysis system. The primary component of the proposed system is an automated Nakagami imaging module with a specialized fuzzy c-means algorithm to contour the boundaries of the whole tumors. The contoured images, afterwards, are analyzed by the Minkowski–Bouligand and Hausdorff methods for two panning options to generate the fractal dimensions and to estimate the fractal complexities for classifying the gliomas The results are greatly encouraging that the overall classification accuracy is computed as 88.31 % using the basic support vector machines (SVM) classifier; while as 91.96 % with the arbitrary thresholding appended. The outcomes of this paper with implementable mathematical infrastructure would be very useful and beneficial as an expert system in intelligent and automatic glioma grading, for researchers and medical experts.

Extraction of Crop Row Navigation Lines for Soybean Seedlings Based on Calculation of Average Pixel Point Coordinates

The extraction of navigation lines is a crucial aspect in the field autopilot system for intelligent agricultural equipment. Given that soybean seedlings are small, and straw can be found in certain Northeast China soybean fields, accurately obtaining feature points and extracting navigation lines during the soybean seedling stage poses numerous challenges. To solve the above problems, this paper proposes a method of extracting navigation lines based on the average coordinate feature points of pixel points in the bean seedling belt according to the calculation of the average coordinate. In this study, the soybean seedling was chosen as the research subject, and the Hue, Saturation, Value (HSV) colour model was employed in conjunction with the maximum interclass variance (OTSU) method for RGB image segmentation. To extract soybean seedling bands, a novel approach of framing binarised image contours by drawing external rectangles and calculating average coordinates of white pixel points as feature points was proposed. The feature points were normalised, and then the improved adaptive DBSCAN clustering method was used to cluster the feature points. The least squares method was used to fit the centre line of the crops and the navigation line, and the results showed that the average distance deviation and the average angle deviation of the proposed algorithm were 7.38 and 0.32. The fitted navigation line achieved an accuracy of 96.77%, meeting the requirements for extracting navigation lines in intelligent agricultural machinery equipment for soybean inter-row cultivation. This provides a theoretical foundation for realising automatic driving of intelligent agricultural machinery in the field.

Research into the Applications of a Multi-Scale Feature Fusion Model in the Recognition of Abnormal Human Behavior.

Due to the increasing severity of aging populations in modern society, the accurate and timely identification of, and responses to, sudden abnormal behaviors of the elderly have become an urgent and important issue. In the current research on computer vision-based abnormal behavior recognition, most algorithms have shown poor generalization and recognition abilities in practical applications, as well as issues with recognizing single actions. To address these problems, an MSCS-DenseNet-LSTM model based on a multi-scale attention mechanism is proposed. This model integrates the MSCS (Multi-Scale Convolutional Structure) module into the initial convolutional layer of the DenseNet model to form a multi-scale convolution structure. It introduces the improved Inception X module into the Dense Block to form an Inception Dense structure, and gradually performs feature fusion through each Dense Block module. The CBAM attention mechanism module is added to the dual-layer LSTM to enhance the model's generalization ability while ensuring the accurate recognition of abnormal actions. Furthermore, to address the issue of single-action abnormal behavior datasets, the RGB image dataset RIDS (RGB image dataset) and the contour image dataset CIDS (contour image dataset) containing various abnormal behaviors were constructed. The experimental results validate that the proposed MSCS-DenseNet-LSTM model achieved an accuracy, sensitivity, and specificity of 98.80%, 98.75%, and 98.82% on the two datasets, and 98.30%, 98.28%, and 98.38%, respectively.

Exploring the Possibility of Automatic Scoring for Arrow Responses of Simple Pendulum Items Using a Image Contours

Exploring the Possibility of Automatic Scoring for Arrow Responses of Simple Pendulum Items Using a Image Contours

A Cortical-Inspired Contour Completion Model Based on Contour Orientation and Thickness.

An extended four-dimensional version of the traditional Petitot-Citti-Sarti model on contour completion in the visual cortex is examined. The neural configuration space is considered as the group of similarity transformations, denoted as M=SIM(2). The left-invariant subbundle of the tangent bundle models possible directions for establishing neural communication. The sub-Riemannian distance is proportional to the energy expended in interneuron activation between two excited border neurons. According to the model, the damaged image contours are restored via sub-Riemannian geodesics in the space M of positions, orientations and thicknesses (scales). We study the geodesic problem in M using geometric control theory techniques. We prove the existence of a minimal geodesic between arbitrary specified boundary conditions. We apply the Pontryagin maximum principle and derive the geodesic equations. In the special cases, we find explicit solutions. In the general case, we provide a qualitative analysis. Finally, we support our model with a simulation of the association field.

Acoustic emission monitoring for damage diagnosis in composite laminates based on deep learning with attention mechanism

During service, advanced composite materials are susceptible to various forms of damage. Developing an efficient real-time Structural Health Monitoring (SHM) method is of profound significance to ensure the normal operational integrity of the advanced composite structures. This study introduces a novel end-to-end SHM approach employing acoustic emission (AE) techniques and a deep learning model with attention mechanisms (AM) to automate the damage diagnosis process and improve prediction accuracy. Low-velocity impact (LVI) experiments were carried out, recording AE signals for subsequent bi-spectrum analysis. Three dedicated AM-based Convolutional Neural Network (CNN) architectures were designed for damage diagnosis, utilizing 2D bi-spectrum contour images as input data. A comparison was made between the proposed CNN model, deep learning and traditional machine learning models in the context of damage diagnostics, revealing that CNN models combined AM exhibit significantly higher accuracy in pattern recognition tasks. The accuracies of the baseline CNN, self-attention-CNN, multi-head attention-CNN and convolutional block attention module-CNN are 98.22%, 98.50%, 98.85%, and 98.87%, respectively. The CNN embedded with AM demonstrated superior predictive performance and fewer instances of misclassification. Considering the specific demands of damage diagnostics in composite materials, the CBAM-CNN is better suited for composite material health monitoring tasks.

Analytical solutions of the space–time fractional Kundu–Eckhaus equation by using modified extended direct algebraic method

The study of soliton solutions for Nonlinear Fractional Partial Differential Equations (NFPDEs) has gained prominence recently because of its ability to realistically recreate complex physical processes. Numerous mathematical techniques have been devised to handle the problem of NFPDEs where soliton solutions are difficult to obtain. Due to their accuracy in reproducing complex physical phenomena, soliton solutions for Nonlinear Fractional Partial Differential Equations (NFPDEs) have recently attracted interest. Several mathematical techniques have been devised to tackle the difficult task of solving non-finite partial differential equations (NFPDEs) soliton. Studies of soliton solutions for Nonlinear Fractional Partial Differential Equations (NFPDEs) have garnered increased attention recently due to its capacity to accurately represent complex physical processes. Due to the difficulty of obtaining soliton solutions, NFPDEs can be solved using a wide variety of mathematical methods. In this way, it facilitates the extraction of the recently found abundance of optical soliton solutions. To further understanding of the results, the study also includes contour and three-dimensional images that visually depict particular optical soliton solutions for particular parameter selections, suggesting the existence of different soliton structures in the nonlinear fractional Kundu–Eckhaus equation (NFKEE) region. It is shown that the proposed technique is quite powerful and effective in solving several nonlinear FDEs.

Pig Weight Estimation Method Based on a Framework Combining Mask R-CNN and Ensemble Regression Model.

Using computer vision technology to estimate pig live weight is an important method to realize pig welfare. But there are two key issues that affect pigs' weight estimation: one is the uneven illumination, which leads to unclear contour extraction of pigs, and the other is the bending of the pig body, which leads to incorrect pig body information. For the first one, Mask R-CNN was used to extract the contour of the pig, and the obtained mask image was converted into a binary image from which we were able to obtain a more accurate contour image. For the second one, the body length, hip width and the distance from the camera to the pig back were corrected by XGBoost and actual measured information. Then we analyzed the rationality of the extracted features. Three feature combination strategies were used to predict pig weight. In total, 1505 back images of 39 pigs obtained using Azure kinect DK were used in the numerical experiments. The highest prediction accuracy is XGBoost, with an MAE of 0.389, RMSE of 0.576, MAPE of 0.318% and R2 of 0.995. We also recommend using the Mask R-CNN + RFR method because it has fairly high precision in each strategy. The experimental results show that our proposed method has excellent performance in live weight estimation of pigs.

Reconstruction of the Chemical Gas Concentration Distribution Using Partial Convolution-Based Image Inpainting.

An interpolation method, which estimates unknown values with constrained information, is based on mathematical calculations. In this study, we addressed interpolation from an image-based perspective and expanded the use of image inpainting to estimate values at unknown points. When chemical gas is dispersed through a chemical attack or terrorism, it is possible to determine the concentration of the gas at each location by utilizing the deployed sensors. By interpolating the concentrations, we can obtain the contours of gas concentration. Accurately distinguishing the contours of a contaminated region from a map enables the optimal response to minimize damage. However, areas with an insufficient number of sensors have less accurate contours than other areas. In order to achieve more accurate contour data, an image inpainting-based method is proposed to enhance reliability by erasing and reconstructing low-accuracy areas in the contour. Partial convolution is used as the machine learning approach for image-inpainting, with the modified loss function for optimization. In order to train the model, we developed a gas diffusion simulation model and generated a gas concentration contour dataset comprising 100,000 contour images. The results of the model were compared to those of Kriging interpolation, one of the conventional spatial interpolation methods, finally demonstrating 13.21% higher accuracy. This suggests that interpolation from an image-based perspective can achieve higher accuracy than numerical interpolation on well-trained data. The proposed method was validated using gas concentration contour data from the verified gas dispersion modeling software Nuclear Biological Chemical Reporting And Modeling System (NBC_RAMS), which was developed by the Agency for Defense Development, South Korea.

AENet: Image Retrieval of Kazakh Handwritten Documents Based on Attention Mechanism and Feature Aggregation

Kazakh is one of the multilingual languages of China and is widely spoken in some areas of Xinjiang, China. However, due to the fact that Kazakh is a language in which several characters are glued together to form a continuous word with a unique shape and complex structural combinations of relationships. This paper explores a solution for offline image retrieval of handwritten Kazakh words, which is a challenging task because, due to the lack of relevant datasets and the special writing morphology of the Kazakh language, traditional text image retrieval algorithms often struggle to achieve satisfactory results when dealing with writing styles that are varied and adherent to the language. Therefore, a dataset of offline Kazakh handwritten document images was created in this paper. The dataset contains 300 pages of document images with 20 500 words. Then, a new model called the ‘AENet’ is proposed. The model utilizes an attention mechanism to focus more finely on focal regions such as centers, inflection points, and contours of handwritten word images and to capture important local features from different scales. Fusion space pyramid pooling, feature aggregation, encoding operations, and feature downscaling and reconstruction are used to extract and reconstruct more representative feature representations from local to global to capture the overall information in the word images. Through experimental evaluation on Kazak‐80, Zilla‐64, and HWDB1.1‐375 datasets, it is verified that the method significantly improves the mAP for image retrieval of handwritten words, which is especially applicable to adhesive languages like Kazakh. © 2024 Institute of Electrical Engineers of Japan and Wiley Periodicals LLC.