Input Image Research Articles

Retinal fundus image super-resolution based on generative adversarial network guided with vascular structure prior

Many ophthalmic and systemic diseases can be screened by analyzing retinal fundus images. The clarity and resolution of retinal fundus images directly determine the effectiveness of clinical diagnosis. Deep learning methods based on generative adversarial networks are used in various research fields due to their powerful generative capabilities, especially image super-resolution. Although Real-ESRGAN is a recently proposed method that excels in processing real-world degraded images, it suffers from structural distortions when super-resolving retinal fundus images are rich in structural information. To address this shortcoming, we first process the input image using a pre-trained U-Net model to obtain a structural segmentation map of the retinal vessels and use the segmentation map as the structural prior. The spatial feature transform layer is then used to better integrate the structural prior into the generation process of the generator. In addition, we introduce channel and spatial attention modules into the skip connections of the discriminator to emphasize meaningful features and accordingly enhance the discriminative power of the discriminator. Based on the original loss functions, we introduce the L1 loss function to measure the pixel-level differences between the segmentation maps of retinal vascular structures in the high-resolution images and the super-resolution images to further constrain the super-resolution images. Simulation results on retinal image datasets show that our improved algorithm results have a better visual performance by suppressing structural distortions in the super-resolution images.

Examining Tongue Movement Intentions in EEG-Based BCI with Machine and Deep Learning: An Approach for Dysphagia Rehabilitation

Abstract Dysphagia, a common swallowing disorder particularly prevalent among older adults and often associated with neurological conditions, significantly affects individuals’ quality of life by negatively impacting their eating habits, physical health, and social interactions. This study investigates the potential of brain-computer interface (BCI) technologies in dysphagia rehabilitation, focusing specifically on motor imagery paradigms based on EEG signals and integration with machine learning and deep learning methods for tongue movement. Traditional machine learning classifiers, such as K-Nearest Neighbors (KNN), Support Vector Machine (SVM), Decision Tree, Naive Bayes, Random Forest, AdaBoost, Bagging, and Kernel were employed in discrimination of rest and imagination phases of EEG signals obtained from 30 healthy subjects. Scalogram images obtained using continuous wavelet transform of EEG signals corresponding to the rest and imagination phases of the experiment were used as the input images to the CNN architecture. As a result, KNN (79.4%) and SVM (63.4%) exhibited lower accuracy rates compared to ensemble methods like AdaBoost, Bagging, and Random Forest, all achieving high accuracy rates of 99.8%. These ensemble techniques proved to be highly effective in handling complex EEG datasets, particularly in distinguishing between rest and imagination phases. Furthermore, the deep learning approach, utilizing CNN and Continuous Wavelet Transform (CWT), achieved an accuracy of 83%, highlighting its potential in analyzing motor imagery data. Overall, this study demonstrates the promising role of BCI technologies and advanced machine learning techniques, especially ensemble and deep learning methods, in improving outcomes for dysphagia rehabilitation.

Image Enhancement Guided Object Detection in Visually Degraded Scenes.

Object detection accuracy degrades seriously in visually degraded scenes. A natural solution is to first enhance the degraded image and then perform object detection. However, it is suboptimal and does not necessarily lead to the improvement of object detection due to the separation of the image enhancement and object detection tasks. To solve this problem, we propose an image enhancement guided object detection method, which refines the detection network with an additional enhancement branch in an end-to-end way. Specifically, the enhancement branch and detection branch are organized in a parallel way, and a feature guided module is designed to connect the two branches, which optimizes the shallow feature of the input image in the detection branch to be as consistent as possible with that of the enhanced image. As the enhancement branch is frozen during training, such a design plays a role in using the features of enhanced images to guide the learning of object detection branch, so as to make the learned detection branch being aware of both image quality and object detection. When testing, the enhancement branch and feature guided module are removed, and so no additional computation cost is introduced for detection. Extensive experimental results, on underwater, hazy, and low-light object detection datasets, demonstrate that the proposed method can improve the detection performance of popular detection networks (YOLO v3, Faster R-CNN, DetectoRS) significantly in visually degraded scenes.

Low‐light image enhancement via lightweight custom non‐linear transform network

AbstractConvolutional neural network (CNN)‐based models have shown significant progress in low light image enhancement. However, many existing models possess a large number of parameters, making them unsuitable for deployment on terminal devices. Moreover, adjustments to brightness, contrast, and colour in images are often non‐linear, and convolution is not the best at capturing complex non‐linear relationships in image data. To address these issues, a model based on an end‐to‐end custom non‐linear transform network (CNTNet) is proposed. CNTNet combines a custom non‐linear transform layer with CNN layers to achieve image contrast and detail enhancement. The CNT layer in this model introduces transformation parameters at multiple scales to manipulate input images within various ranges. CNTNet progressively processes images by stacking multiple non‐linear transform layers and convolutional layers while integrating residual connections to capture and leverage subtle image features. The final output is generated through convolutional layers to obtain enhanced images. Experimental results of CNTNet demonstrate that, while maintaining a comparable level of image quality evaluation metrics to mainstream models, it significantly reduces the parameter count to only 2K.

Machine/deep learning-assisted hemoglobin level prediction using palpebral conjunctival images.

Palpebral conjunctival hue alteration is used in non-invasive screening for anaemia, whereas it is a qualitative measure. This study constructed machine/deep learning models for predicting haemoglobin values using 150 palpebral conjunctival images taken by a smartphone. The median haemoglobin value was 13.1 g/dL, including 10 patients with <11 g/dL. A segmentation model using U-net was successfully constructed. The segmented images were subjected to non-convolutional neural network (CNN)-based and CNN-based regression models for predicting haemoglobin values. The correlation coefficients between the actual and predicted haemoglobin values were 0.38 and 0.44 in the non-CNN-based and CNN-based models, respectively. The sensitivity and specificity for anaemia detection were 13% and 98% for the non-CNN-based model and 20% and 99% for the CNN-based model. The performance of the CNN-based model did not improve with a mask layer guiding the model's attention towards the conjunctival regions, however, slightly improved with correction by the aspect ratio and exposure time of input images. The gradient-weighted class activation mapping heatmap indicated that the lower half area of the conjunctiva was crucial for haemoglobin value prediction. In conclusion, the CNN-based model had better results than the non-CNN-based model. The prediction accuracy would improve by using more input data with anaemia.

Lightweight cross-modal transformer for RGB-D salient object detection

Recently, Transformer-based RGB-D salient object detection (SOD) models have pushed the performance to a new level. However, they come at the cost of consuming abundant resources, including memory and power, thus hindering their real-life applications. To remedy this situation, a novel lightweight cross-modal Transformer (LCT) for RGB-D SOD will be presented in this paper. Specifically, LCT will first reduce its parameters and computational costs by employing a middle-level feature fusion structure and taking a lightweight Transformer as the backbone. Then, with the aid of Transformers, it will compensate for performance degradation by effectively capturing the cross-modal and cross-level complementary information from the multi-modal input images. To this end, a cross-modal enhancement and fusion module (CEFM) with a lightweight channel-wise cross attention block (LCCAB) will be designed to capture the cross-modal complementary information effectively but with fewer costs. A bi-directional multi-level feature interaction module (Bi-MFIM) with a lightweight spatial-wise cross attention block (LSCAB) will be designed to capture the cross-level complementary context information. By virtue of CEFM and Bi-MFIM, the performance degradation caused by parameter reduction can be well compensated, thus boosting the performances. By doing so, our proposed model has only 2.8M parameters with 7.6G FLOPs and runs at 66 FPS. Furthermore, experimental results on several benchmark datasets show that our proposed model can achieve competitive or even better results than other models. Our code will be released on https://github.com/nexiakele/lightweight-cross-modal-Transformer-LCT-for-RGB-D-SOD.

Detection of artificial spots in fundus images using modified U-Net based semantic segmentation

Today, deep learning algorithms are playing a very crucial role in the initial stage diagnosis of several fundus diseases like glaucoma, hypertension, and diabetic retinopathy. Lots of research is going on in this area now-a-days. During the acquisition of fundus images some artificial spots (e.g. because of the device itself, dust particles in the surroundings) have been added in captured images. In this paper, artificial spots in the fundus images that are generated due to the non-standardized conditions of scanning devices, are detected with the help of a newly proposed modified UNet (mU-Net) semantic segmentation model. Initially, preprocessing methods such as Gaussian blur, thresholding, and Hough transform have been used to create artificial spots. Now these preprocessed images have been used for training the proposed model. To make the proposed model more effective, the following modifications like, Regularization techniques (early stopping, greater weight decay, and Adam optimizer), decay learning rate scheduler, categorical cross-entropy loss function, and a significant number of filters have been modified in simple U-Net model. Apart from these mentioned modifications in the base U-Net model, a feature injecting module (FIM) has been added between the expansion and the contraction section of the simple U-Net model. FIM adds the features of the input image at the time of up-sampling. The addition of FIM to a simple U-Net model improves the detection of artificial spots and enhances the performance of the model. The mU-Net has been compared with other models, namely simple U-Net, V-Net, UNet++, ResUnet-a, WideU-Net, and Swin-Unet. The Friedman test that has been conducted on IOU, DICE, MAE, PSNR, and SSIM scores, found that mU-Net balances evaluation metrics well. It appears that the nonparametric Friedman test will improve reproducibility by demonstrating statistical significance. The IOU, DICE, MAE, PSNR, and SSIM scores of the proposed model indicate superior performance as compared to other models.

EcD-Net: Encoder-Corollary Atrous Spatial Pyramid Pooling-Decoder Network for Automated Pancreas Segmentation of 2D CT Images

Automatic pancreas segmentation of CT scans enables physicians to identify and monitor the abnormalities in the pancreas. This facilitates intraoperative assistance, surgical planning, prognosis, and diagnosis. Nonetheless, the size and location of the pancreas in the CT image input data present a significant challenge for automatic segmentation, and the intricacy of the background region confounds deep segmentation networks. To solve this difficulty, we propose the Encoder-Corollary Atrous Spatial Pyramid Pooling-Decoder Network (EcD-Net) for locating and segmenting the pancreas. This two-tiered method begins with a coarse segmentation stage for locating the pancreas within the overall CT image. Using the detected image from the first tier, a fine segmentation network based on U-Net is applied to segment the target organ (pancreas). A novel Saturated Multi-view Dense Module (SMD- Module) is presented to enhance information gradient flow for the stability of the training process and easier convergence at the fine stage. A novel Corollary Atrous Spatial Pyramid Pooling Module (CASPP-Module) is proposed to simultaneously capture low-level details and high-level global context information to enhance the pancreas segmentation accuracy, extract global and local spatial data, and detect the pancreas at varied scales. On the publicly available National Institute of Health (NIH) pancreas dataset, our proposed EcD-Net surpasses previous state-of-the-art methods with a mean Dice Similarity Coefficient (DSC) of 88.84%, mean Precision (PRE), Recall (REC), and Intersection over Union (IoU) of 91.85%, 88.21%, and 89.94% respectively, and it also records the lowest standard deviations showing the robustness and steadiness of our proposed method.

Automatic thresholding method using single information entropy under product transformation of order difference filter response

To automatically threshold images with unimodal, bimodal, multimodal or non-modal gray level distributions within a unified framework, an automatic thresholding method using single information entropy under the product transformation of order difference filter response is proposed. The proposed method first performs the product transformation of order difference filter response on an input image at different scales to obtain the product transformation image. Critical or non-critical pixels are labelled on each pixel of the binary images corresponding to different thresholds to construct a series of binary label images that are used for distinguishing critical or non-critical regions. A single information entropy is finally used for characterizing the information obtained from the product transformation image with the critical regions of different binary label images, and the threshold corresponding to maximum information entropy is selected as final threshold. The proposed method is compared with seven state-of-the-art segmentation methods. Experimental results on 12 synthetic images and 98 real-world images show that the average Matthews correlation coefficients of the proposed method reached 0.994 and 0.966 for the synthetic images and the real-world images, which outperform the second-best method by 52.4 % and 27.8 %, respectively. The proposed method has more robust segmentation adaptability to test images with different modalities, despite not offering an advantage in terms of computational efficiency.

Reconstructing and analyzing the invariances of low-dose CT image denoising networks.

Deep learning-based methods led to significant advancements in many areas of medical imaging, most of which are concerned with the reduction of artifacts caused by motion, scatter, or noise. However, with most neural networks being black boxes, they remain notoriously difficult to interpret, hindering their clinical implementation. In particular, it has been shown that networks exhibit invariances w.r.t. input features, that is, they learn to ignore certain information in the inputdata. To improve the interpretability of deep learning-based low-dose CT image denoisingnetworks. We learn a complete data representation of low-dose input images using a conditional variational autoencoder (cVAE). In this representation, invariances of any given denoising network are then disentangled from the information it is not invariant to using a conditional invertible neural network (cINN). At test time, image-space invariances are generated by applying the inverse of the cINN and subsequent decoding using the cVAE. We propose two methods to analyze sampled invariances and to find those that correspond to alterations of anatomicalstructures. The proposed method is applied to four popular deep learning-based low-dose CT image denoising networks. We find that the networks are not only invariant to noise amplitude and realizations, but also to anatomicalstructures. The proposed method is capable of reconstructing and analyzing invariances of deep learning-based low-dose CT image denoising networks. This is an important step toward interpreting deep learning-based methods for medical imaging, which is essential for their clinical implementation.

A Deep Learning-based Approach for Channel Estimation in Multi-access Multi-antenna Systems

This paper studies estimating the channel state information at the end of receiver (CSIR) for multiple transmitters communicating with only one receiver so that the latter can decode the incoming signal more efficiently. The transmitters and the receiver are all equipped with multi-antennas and using orthogonal space-time block codes (OSTBC). An algorithm is developed based on deep learning for estimating multi-user multiple-input multiple-output (MU-MIMO) channels. The algorithm could estimate the CSIR using a single pilot block. The proposed convolutional neural network (CNN) architecture designed for this task begins with an input layer that accepts grayscale images, followed by six convolutional blocks for feature extraction and processing. The network concludes with a fully connected layer to output the estimated channel information. It is trained using a regression loss function to map input images to accurate channel information accurately. The performance of the proposed method is compared with classical methods like least square and subspace-based methods, including Capon and rank revealing QR (RRQR) methods. CNN achieved better performance in comparison with the reference. Computer simulations are included to validate the proposed method.

Autonomous Single-Image Dehazing: Enhancing Local Texture with Haze Density-Aware Image Blending

Single-image dehazing is an ill-posed problem that has attracted a myriad of research efforts. However, virtually all methods proposed thus far assume that input images are already affected by haze. Little effort has been spent on autonomous single-image dehazing. Even though deep learning dehazing models, with their widely claimed attribute of generalizability, do not exhibit satisfactory performance on images with various haze conditions. In this paper, we present a novel approach for autonomous single-image dehazing. Our approach consists of four major steps: sharpness enhancement, adaptive dehazing, image blending, and adaptive tone remapping. A global haze density weight drives the adaptive dehazing and tone remapping to handle images with various haze conditions, including those that are haze-free or affected by mild, moderate, and dense haze. Meanwhile, the proposed approach adopts patch-based haze density weights to guide the image blending, resulting in enhanced local texture. Comparative performance analysis with state-of-the-art methods demonstrates the efficacy of our proposed approach.

Region‐selective shift register by a single output enable signal for a multiple refresh frequency driving scheme

AbstractMany low‐power technologies have been developed for high‐performance displays. One method is to vary the refresh rate according to the input image, where frequencies are reduced over still images and slow‐moving videos. To support the variable refresh rate scheme, the programmable gate driver circuit should be integrated. This paper proposes a thin‐film transistor (TFT) shift register of which output pulse is simply controlled by an enable signal. Because the output pulse generation is directly adjusted without the previous programming period, various output pulse sequences are possible with the high flexibility. The shift register is composed by adding six TFTs and one capacitor to a conventional circuit that provides a carry pulse to a next stage. The simulation program with integrated circuit emphasis (SPICE) simulation ensures that gate pulses for any numbers of lines in any regions can be generated within one frame only by applying the enable signal at the corresponding timing, resulting in multiple refresh frequencies in a screen. In addition, it is verified that additional TFTs and capacitor contribute to only the negligible increase in power consumption by 2.1%, compared to a conventional shift register.

Dual phase hierarchical brain tumour detection and segmentation using unet based skip guidance residual convolutional transformer

Brain tumour segmentation and classification is an important approach to understanding tumour characteristics effectively. Accurate lesion segmentation requires many image modalities with varied contrasts. As a result, manual segmentation may not be suitable for large-scale research. The research has shown interest in developing various learning algorithms to detect and segment brain tumour (BT) images accurately based on the set of experimental images. Some of the issues noticed in the existing models, such as high computational complexity, high processing time, and less accuracy, were identified. In order to overcome these existing issues, a novel Dual-phase channel attentional Residual convolutional transformer Unet model (DCARCTUnet) is introduced for segmenting brain tumours. Initially, the input images are passed into the pre-processing stage to remove noise and blur. The Efficient Net model identified the pre-processed image as either normal or tumour. The proposed segmentation model can be integrated into three stages, namely dual phase channel attention module (DCAM), residual convolutional transformer Unet model and Pyramid skip guidance module (PSM). The proposed method uses the BRATS 2020–2021 dataset to analyze the results accurately. The results showed that the proposed model has a dice coefficient value of 0.99 and a Jaccard coefficient value of 0.98. Furthermore, performance analysis is carried out to demonstrate the effectiveness of the proposed model.

Feature-based detection of breast cancer using convolutional neural network and feature engineering.

Breast cancer (BC) is a prominent cause of female mortality on a global scale. Recently, there has been growing interest in utilizing blood and tissue-based biomarkers to detect and diagnose BC, as this method offers a non-invasive approach. To improve the classification and prediction of BC using large biomarker datasets, several machine-learning techniques have been proposed. In this paper, we present a multi-stage approach that consists of computing new features and then sorting them into an input image for the ResNet50 neural network. The method involves transforming the original values into normalized values based on their membership in the Gaussian distribution of healthy and BC samples of each feature. To test the effectiveness of our proposed approach, we employed the Coimbra and Wisconsin datasets. The results demonstrate efficient performance improvement, with an accuracy of 100% and 100% using the Coimbra and Wisconsin datasets, respectively. Furthermore, the comparison with existing literature validates the reliability and effectiveness of our methodology, where the normalized value can reduce the misclassified samples of ML techniques because of its generality.

Security of End-to-End medical images encryption system using trained deep learning encryption and decryption network

The Internet of Medical Things (IoMT) links medical devices and wearable, enhancing healthcare. To secure sensitive patient data over the IoMT, encryption is vital to retain confidentiality, prevent tampering, ensure authenticity, and secure data transfer. The intricate neural network architecture of deep learning models adds a layer of complexity and non-linearity to the encryption process, rendering it highly resistant to plaintext attacks. Specifically, the Cycle_GAN network is used as the leading learning network. This work suggests deep learning-based encryption for medical images using Cycle_GAN, a Generative Adversarial Network. Cycle_GAN changes images without paired training data that improves quality and feature preservation. Unlike conventional image-to-image translation techniques, Cycle_GAN doesn’t require a dataset with corresponding input–output pairs. Traditional methods typically needs paired data to learn the mapping between input and output images. Paired data can be challenging to obtain, specifically in medical imaging where gathering and annotating data can be time-consuming, laborious and expensive. The use of Cycle_GAN overwhelms this constraint by using unpaired data, where the input and output images are not explicitly paired. This method ensures confidentiality, authenticity, and secure transfer. Cycle_GAN consists of two major components: a generator used to modify the images, and a discriminator used to distinguish between real and fake images. Further, the Binary-Cross Entropy loss function is employed to train the network for precise predictions. The experiments are carried out on skin cancer datasets. The results demonstrate high-level efficient, systematic and coherent encryption as compared with other modernized medical image encryption methods. The proposed technique offers several benefits, including efficient encryption and decryption and robustness against unauthorized access.

Improving Re-Identification by Estimating and Utilizing Diverse Uncertainty Types for Embeddings

In most re-identification approaches, embedding vectors are compared to identify the best match for a given query. However, this comparison does not take into account whether the encoded information in the embedding vectors was extracted reliably from the input images. We propose the first attempt that illustrates how all three types of uncertainty, namely model uncertainty (also known as epistemic uncertainty), data uncertainty (also known as aleatoric uncertainty), and distributional uncertainty, can be estimated for embedding vectors. We provide evidence that we do indeed estimate these types of uncertainty, and that each type has its own value for improving re-identification performance. In particular, while the few state-of-the-art approaches that employ uncertainty for re-identification during inference utilize only data uncertainty to improve single-shot re-identification performance, we demonstrate that the estimated model uncertainty vector can be utilized to modify the feature vector. We explore the best method for utilizing the estimated model uncertainty based on the Market-1501 dataset and demonstrate that we are able to further enhance the performance above the already strong baseline UAL. Additionally, we show that the estimated distributional uncertainty resembles the degree to which the current sample is out-of-distribution. To illustrate this, we divide the distractor set of the Market-1501 dataset into four classes, each representing a different degree of out-of-distribution. By computing a score based on the estimated distributional uncertainty vector, we are able to correctly order the four distractor classes and to differentiate them from an in-distribution set to a significant extent.

Three-stage binarization of color document images based on discrete wavelet transform and generative adversarial networks

The efficient extraction of text information from the background in degraded color document images is an important challenge in the preservation of ancient manuscripts. The imperfect preservation of ancient manuscripts has led to different types of degradation over time, such as page yellowing, staining, and ink bleeding, seriously affecting the results of document image binarization. This work proposes an effective three-stage network method to image enhancement and binarization of degraded documents using generative adversarial networks (GANs). Specifically, in Stage-1, we first split the input images into multiple patches, and then split these patches into four single-channel patch images (gray, red, green, and blue). Then, three single-channel patch images (red, green, and blue) are processed by the discrete wavelet transform (DWT) with normalization. In Stage-2, we use four independent generators to separately train GAN models based on the four channels on the processed patch images to extract color foreground information. Finally, in Stage-3, we train two independent GAN models on the outputs of Stage-2 and the resized original input images (512 × 512) as the local and global predictions to obtain the final outputs. The experimental results show that the Avg-Score metrics of the proposed method are 77.64, 77.95, 79.05, 76.38, 75.34, and 77.00 on the (H)-DIBCO 2011, 2013, 2014, 2016, 2017, and 2018 datasets, which are at the state-of-the-art level. The implementation code for this work is available at https://github.com/abcpp12383/ThreeStageBinarization.

Quaternion attention-based JND model for macrophotography image watermarking

With the advancement of imaging technology, macrophotography images (MPIs) have become a popular research topic. Unlike natural images, MPIs often feature sharp foregrounds and blurred backgrounds, leading to distinct perceptual characteristics in estimation. As the number of MPIs grows rapidly, concerns over image quality and security increase. Robust watermarking techniques have been introduced to address these challenges. Just Noticeable Difference (JND) has been widely used in quantization-based watermarking frameworks. However, existing JND models handle each image area with a single-level perceptual attention. Visual attention in Quaternion Discrete Wavelet Transform (QDWT), which can reflect the Multi-level perceptual attention feature. Therefore, we propose a new method called Quaternion Attention-based Just Noticeable Difference model for MPIs Watemarking (QAJnd-MW) for watermarking MPIs. This method uses visual attention mechanisms, recognizing that the HVS is more sensitive to attention regions. We generate a masking effect in the JND field. The input image undergoes QDWT to explore multi-scale features. The multi-scale feature maps, with multi-directional luminance and multi-channel color, help create local and global attention maps, which are fused to form the final attention map. Specifically, considering both attention-based masking effects, the quaternion attention-guided JND model is designed for a robust MPI watermarking framework, aiming to further improve MPI security. Extensive experiments on the MP2020 and Blur Detection datasets show that the proposed model significantly improves robustness against JPEG compression attacks, reducing the bit error rate (BER) by up to 12%. Additionally, the model performs well against other attacks, such as those in online social networks, with lower BER than current state-of-the-art techniques.

License Plate Recognition Using Three-Dimensional Rotated Character Recognition and Instance Segmentation by Deep Learning

License plate recognition is currently used in various situations, such as parking lot vehicle management and the tracking of wanted vehicles. These applications require recognition from all possible camera angles. Therefore, we propose a method for license plate recognition in images captured from various camera angles. First, the license plate area is detected from the input image using the YOLOv5 object detection method. The subsequent process can be performed in two manners: by rotating the license plate to the front using projective transformation and trapezoidal correction, or by using graph matching. The first method performs Hough transforms or linear approximations as preprocessing, and then calculates the rectangle surrounding the license plate from the YOLOv5 output. After performing trapezoidal correction, the maximally stable extremal regions (MSER) are used to detect character candidates, and characters are recognized using 3D rotated character recognition. The second method performs character candidate detection and recognition without trapezoidal correction, followed by graph matching. These methods are versatile because they do not depend on the layout of license plates, which varies among countries and regions. Two types of datasets were used: a set of images containing Japanese license plates collected by ourselves, and a set of publicly available images containing Taiwanese license plates. Two recognition rates were evaluated: the license plate character recognition rate and license plate recognition rate in which all characters in one plate are recognized. The license plate character recognition rates using graph matching were 91.9% for the Japanese license plates and 89.3% for the Taiwanese license plates. The license plate recognition rates were 84.0% and 84.5%, respectively.