Blood vessel segmentation in eye fundus images

The automatic analysis of retinal blood vessels plays an important role in the computer-aided diagnosis. In this paper, we introduce a probabilistic tracking-based method for automatic vessel segmentation in retinal images. We take into account vessel edge detection on the whole retinal image and handle different vessel structures. During the tracking process, a Bayesian method with maximum a posteriori (MAP) as criterion is used to detect vessel edge points. Experimental evaluations of the tracking algorithm are performed on real retinal images from three publicly available databases: STARE (Hoover et al., 2000), DRIVE (Staal et al., 2004), and REVIEW (Al-Diri et al., 2008 and 2009). We got high accuracy in vessel segmentation, width measurements, and vessel structure identification. The sensitivity and specificity on STARE are 0.7248 and 0.9666, respectively. On DRIVE, the sensitivity is 0.6522 and the specificity is up to 0.9710.

Automated blood vessel segmentation of retinal images offers huge potential benefits for medical diagnosis of different ocular diseases. In this paper, 2D Matched Filters (MF) are applied to fundus retinal images to detect vessels which are enhanced by Contrast Limited Adaptive Histogram Equalization (CLAHE) method. Due to the Gaussian nature of blood vessel profile, the MF with Gaussian kernel often misclassifies non-vascular structures (e.g., step, ramp or other transients) as vessels. To avoid such false detection, this paper introduces Laplacian of Gaussian (LoG) filters in the vessel segmentation process. The inherent zero-crossing property of LoG filter is used in the algorithm, along with the MF, in order to extract vessels reliably from the retinal images. The proposed method is validated against three publicly available databases, STARE, DRIVE and HRF. Simulation results show that the proposed method is able to segment vessels accurately from the three database images with an average accuracy that is competitive to the existing methodologies.

Vessel segmentation of retinal images is a key diagnostic capability in ophthalmology. This problem faces several challenges including low contrast, variable vessel size and thickness, and presence of interfering pathology such as micro-aneurysms and hemorrhages. Early approaches addressing this problem employed hand-crafted filters to capture vessel structures, accompanied by morphological post-processing. More recently, deep learning techniques have been employed with significantly enhanced segmentation accuracy. We propose a novel domain enriched deep network that consists of two components: 1) a representation network that learns geometric features specific to retinal images, and 2) a custom designed computationally efficient residual task network that utilizes the features obtained from the representation layer to perform pixel-level segmentation. The representation and task networks are jointly learned for any given training set. To obtain physically meaningful and practically effective representation filters, we propose two new constraints that are inspired by expected prior structure on these filters: 1) orientation constraint that promotes geometric diversity of curvilinear features, and 2) a data adaptive noise regularizer that penalizes false positives. Multi-scale extensions are developed to enable accurate detection of thin vessels. Experiments performed on three challenging benchmark databases under a variety of training scenarios show that the proposed prior guided deep network outperforms state of the art alternatives as measured by common evaluation metrics, while being more economical in network size and inference time.

Accurate vessel segmentation in retinal images is vital for retinopathy diagnosis and analysis. However, existence of very thin vessels in low image contrast along with pathological conditions (e.g., capillary dilation or microaneurysms) render the segmentation task difficult. In this work, we present a novel approach for retinal vessel segmentation focusing on improving thin vessel segmentation. We develop a deep convolutional neural network (CNN), which exploits the specific characteristics of the input retinal data to use deep supervision, for improved segmentation accuracy. In particular, we use the average input retinal vessel width and match it with the layer-wise effective receptive fields (LERF) of the CNN to determine the location of the auxiliary supervision. This helps the network to pay more attention to thin vessels, that otherwise the network would ‘ignore’ during training. We verify our method on three public retinal vessel segmentation datasets (DRIVE, CHASE_DB1, and STARE), achieving better sensitivity (10.18% average increase) than state-of-the-art methods while maintaining comparable specificity, accuracy, and AUC.

Vessel segmentation is an important problem in medical image analysis and is often challenging due to large variations in vessel appearance and profiles, as well as image noises. To address these challenges, we propose a solution by combining heterogeneous context-aware features with a discriminative learning framework. Our solution is characterized by three key ingredients: First, we design a hybrid feature pool containing recently invented descriptors including the stroke width transform (SWT) and Weber's local descriptors (WLD), as well as classical local features including intensity values, Gabor responses and vesselness measurements. Second, we encode context information by sampling the hybrid features from an orientation invariant local context. Third, we treat pixel-level vessel segmentation as a discriminative classification problem, and use a random forest to fuse the rich information encoded in the hybrid context-aware features. For evaluation, the proposed method is applied to retinal vessel segmentation using three publicly available benchmark datasets. On the DRIVE and STARE datasets, our approach achieves average classification accuracies of 0.9474 and 0.9633, respectively. On the high-resolution dataset HRFID, our approach achieves average classification accuracies of 0.9647, 0.9561 and 0.9634 on three different categories, respectively. Experiments are also conducted to validate the superiority of hybrid feature fusion over each individual component.

Fire outbreaks continue to cause damage despite the improvements in fire-detection tools and algorithms. As the human population and global warming continue to rise, fires have emerged as a significant worldwide issue. These factors may contribute to the greenhouse effect and climatic changes, among other detrimental consequences. It is still challenging to implement a well-performing and optimized approach, which is sufficiently accurate, and has tractable complexity and a low false alarm rate. A small fire and the identification of a fire from a long distance are also challenges in previously proposed techniques. In this study, we propose a novel hybrid model, called IS-CNN-LSTM, based on convolutional neural networks (CNN) to detect and analyze fire intensity. A total of 21 convolutional layers, 24 rectified linear unit (ReLU) layers, 6 pooling layers, 3 fully connected layers, 2 dropout layers, and a softmax layer are included in the proposed 57-layer CNN model. Our proposed model performs instance segmentation to distinguish between fire and non-fire events. To reduce the intricacy of the proposed model, we also propose a key-frame extraction algorithm. The proposed model uses Internet of Things (IoT) devices to alert the relevant person by calculating the severity of the fire. Our proposed model is tested on a publicly available dataset having fire and normal videos. The achievement of 95.25% classification accuracy, 0.09% false positive rate (FPR), 0.65% false negative rate (FNR), and a prediction time of 0.08 s validates the proposed system.

Optic Disc (OD) localization is an important step in retinal image analysis especially for grading Diabetic retinopathy and Glaucoma. This paper describes a method to automatically localize the OD position in retinal fundus images. The method localizes the OD by combining intensity and vascular information. Initially, the vascular structure is extracted using 2 dimensional Gabor filtering followed by a simple thresholding technique. Then the initial OD candidates are selected using intensity information. The candidates’ satisfying area and density criterion are considered for the final stage. In the final stage, the entropy of vascular directions on the final candidates was calculated, and the candidate having maximum and optimum entropy was considered as OD region. The proposed method was evaluated on the two publicly available DRIVE and STARE databases. The method was able to obtain 100% of OD localization accuracy on DRIVE database with 2.2 seconds average computation time.

Detecting curbs efficiently and at a low cost is essential for autonomous driving. Most of the existing methods rely on light detection and ranging, which is a complicated and costly solution. Therefore, this work propose an efficient method for detecting curbs in a single road image. This method utilizes the convolutional neural network (CNN) framework to localize the curb parts in the captured image. We built a feature-extraction CNN that can classify curb and no-curb regions. The relu, pooling, fully connected, softmax, and classification layers in the pretrained ResNet18 are replaced with ten new layers, which are relu, convolution2d, batchnormalization, dropout, fully connected, relu, maxpooling2d, fully connected, softmax, and classification layers. The comparison of several evaluation metrics to prove the effectiveness of the customized network structure for curb classification. The customized network structure is assembled in the Faster R-CNN detector to improve the performance of feature extraction both for classification and detection. In the detection stage, although the layer we used for feature extraction was the previous layer of the added layers, the detection experimental results demonstrate that our custom CNN network structure can improve the classification performance of the feature extraction layer.

The application of deep learning technology has increased rapidly in recent years. Technologies in deep learning increasingly emulate natural human abilities, such as knowledge learning, problem-solving, and decision-making. In general, deep learning can carry out self-training without repetitive programming by humans. Convolutional neural networks (CNNs) are deep learning algorithms commonly used in wide applications. CNN is often used for image classification, segmentation, object detection, video processing, natural language processing, and speech recognition. CNN has four layers: convolution layer, pooling layer, fully connected layer, and non-linear layer. The convolutional layer uses kernel filters to calculate the convolution of the input image by extracting the fundamental features. The pooling layer combines two successive convolutional layers. The third layer is the fully connected layer, commonly called the convolutional output layer. The activation function defines the output of a neural network, such as 'yes' or 'no'. The most common and popular CNN activation functions are Sigmoid, Tanh, ReLU, Leaky ReLU, Noisy ReLU, and Parametric Linear Units. The organization and function of the visual cortex greatly influence CNN architecture because it is designed to resemble the neuronal connections in the human brain. Some of the popular CNN architectures are LeNet, AlexNet and VGGNet.

A wide variety of security applications like computers, laptops, mobile phones, ATM machines, physical access control to property etc. are using biometrics based system for human identification. Today’s Biometrics authentication system can provide highest level of security to the user of the service and the service providers. Today’s uni-modal biometric systems has become mature system and being used for low level to medium level security applications. But, for extremely higher level of security requirements like military base, nuclear plants, research laboratories etc., Finger knuckle image and Retina image based combination can be used, as this combinations provides the highest level of accuracy, low error rates, universality etc. Also the security of Multi-modal biometric templates is very importance as biometric template store vital information of the user. In this research work, Multi-modal Biometric system for higher level security using Finger Knuckle image and Retina image with templates security is developed, experiments are carried out and results are calculated using PolyU database for Finger Knuckle image and DRIVE for Retina images. A Finger Knuckle Image processing starts with Median Filtering, CLAHE method and Canny edge detection. The features extraction is carried using SURF feature method. Once the finger knuckle image template is generated, it is encrypted using RSA algorithm and stored in database. The Retina Image Processing is carried out by applying Green channel followed by Gaussian filtering, CLAHE method is applied on outcome of Gaussian filtering. SURF features set is extracted from retina ROI. Retina image template is generated which is later encrypted using RSA algorithm. The final experiment shows that genuine acceptance rate = 100% is achieved for threshold value 0.35 and False Acceptance Rate is only 1%. The scientific contribution of proposed research work is a multi-biometric system with higher accuracy (GAR is 100%), low error rate (FAR is 1%), better template protection by RSA encryption and efficient hybrid model using finger knuckle image and retina image.

In this paper, we have developed a new method of accurate detection of retinal blood vessels based on a deep convolutional neural network (CNN) model. This method plays an important role in the observation of many eye diseases. Retinal Images have many issues that make the process of vessels segmentation very hard. We treat each issue of the retina image with the greatest observation to obtain a well-segmented image. The first step is to apply a pre-processing method based on fuzzy logic and image processing tactics. In a second step, in order to generate the segmented images, we propose a strided encoder-decoder CNN model. This network is trained and optimized using the Dice Loss function that supports the class imbalance problem that is in the database. The proposed model has a U-Net shape, but it is deeper and the pooling layers are replaced with strided convolutional layers in the encoder. This modification allows for a more precise segmentation of vessels and accelerates the training process. The last step is post-processing for removing the noisy pixels as well as the shadow of the optic disc. The performance of the proposed method was evaluated on DRIVE and STARE databases. The proposed method gives a sensitivity of 0.802 and 0.801 respectively on DRIVE and STARE, with an accuracy of 0.959 and 0.961 respectively. We focused on sensitivity and accuracy measurements that represent the accuracy of the model, especially tiny vessels. According to the results, the model outperforms many other proposed methods, especially in the above-mentioned measures.

Computer-aided detection (CAD) provides an efficient way to assist doctors to interpret fundus images. In a CAD system, retinal vessel (RV) detection is an important step to identify the retinal disease regions automatically and accurately. However, RV detection is still a challenging problem due to variations in morphology of the vessels on a noisy background. In this paper, we formulate the detection task as a classification problem and solve it using a convolutional neural network (CNN) as a two-class classifier. The proposed model has 2 convolution layers, 2 pooling layers, 1 dropout layer and 1 loss layer. The proposed CNN achieves better performance and significantly outperforms the state-of-the-art for automatic retinal vessel segmentation on the DRIVE dataset with 91.78% accuracy and 0.96743 AUC score. We further compare our result with several state of the art methods based on AUC values. The comparison shows that our proposal yields the second best AUC value. This demonstrates the efficiency of the proposed method which has no pre-processing steps.

This paper develops a wearable sport activity classification system and its associated deep learning-based sport activity classification algorithm for accurately recognizing sport activities. The proposed wearable system used two wearable inertial sensing modules worn on athletes’ wrist and ankle to collect sport motion signals and utilized a deep convolutional neural network (CNN) to extract the inherent features from the spectrograms of the short-term Fourier transform (STFT) of the sport motion signals. The wearable inertial sensing module is composed of a microcontroller, a triaxial accelerometer, a triaxial gyroscope, an RF wireless transmission module, and a power supply circuit. All ten participants wore the two wearable inertial sensing modules on their wrist and ankle to collect motion signals generated by sport activities. Subsequently, we developed a deep learning-based sport activity classification algorithm composed of sport motion signal collection, signal preprocessing, sport motion segmentation, signal normalization, spectrogram generation, image mergence/resizing, and CNN-based classification to recognize ten types of sport activities. The CNN classifier consisting of two convolutional layers, two pooling layers, a fully-connected layer, and a softmax layer can be used to divide the sport activities into table tennis, tennis, badminton, golf, batting baseball, shooting basketball, volleyball, dribbling basketball, running, and bicycling, respectively. Finally, the experimental results show that the proposed wearable sport activity classification system and its deep learning-based sport activity classification algorithm can recognize 10 sport activities with the classification rate of 99.30%.

In the steel industry, billet numbers are typically identified before the rolling process because of customers’ different requirements. To identify the billet number, computer vision systems are widely used, because the billet number is marked on the front of the billet by a specialized marking machine at a high temperature. Conventional algorithms, such as rule-based and machine-learning-based algorithms, require features of objects. The features are designed by the user, and they significantly influence the accuracy of the algorithm. To address this problem, deep learning methods have recently been researched. In image processing, the convolutional neural network (CNN) is widely used among the deep learning methods. We propose an end-to-end algorithm using CNN to detect and recognize the billet numbers used in the steel industry. The proposed algorithm consists of four convolutional layers, three pooling layers, two fully connected layers, and a softmax layer. The output of the CNN model consists of the probabilistic values for 18 classes, which include 17 character classes and 1 background class. By using the output of the CNN model, we obtain a character confidence matrix, and by using the score function, the optimal position of the billet number is detected, and the optimal character is classified. Furthermore, we exploit the fact that the billet number consists of four columns and two rows. The experimental results show that the billet number recognition accuracy is maximized as 95.1% in 20 epochs. Using the proposed algorithm will help to increase operation efficiency in the steel industry.

In this paper we present time analysis results of speech emotion recognition using convolutional neural network architecture and spectrograms as a speech features. Analyses were performed on model with two convolutional layers followed by pooling layer, and one fully-connected layer followed by dropout and softmax layer on the output. On this model we analyzed time characteristics of speech signal represented by spectrograms. The aim of our work was to find relation between duration of speech signal and the recognition rate of seven basic emotions. It was discovered that speech length is important and naturally the accuracy is growing with the length of analyzed window, however over approximately 1.2 seconds the growth becomes rather mild.