This chapter contains sections titled: Run-Length Coding Binary Tree and Quadtree Contour Representation Skeletonization by Thinning Medial Axis Transformation Object Representation and Tolerance References ]]>

Land cover segmentation has been a significant research area because of its multiple applications including the infrastructure development, forestry, agriculture, urban planning, and climate change research. In this paper, we propose a novel segmentation method, called Frequency-guided Position-based Attention Network (FPA-Net), for land cover image segmentation. Our method is based on encoder–decoder improved U-Net architecture with position-based attention mechanism and frequency-guided component. The position-based attention block is used to capture the spatial dependency among different feature maps and obtain the relationship among relevant patterns across the image. The frequency-guided component provides additional support with high-frequency features. Our model is simple and efficient in terms of time and space complexities. Experimental results on the Deep Globe, GID-15, and Land Cover AI datasets show that the proposed FPA-Net can achieve the best performance in both quantitative and qualitative measures as compared against other existing approaches.

Value-based algorithms have achieved great successes in solving Reinforcement Learning problems via minimizing the mean squared Bellman error (MSBE). Temporal-difference (TD) algorithms such as Q-learning and SARSA often use stochastic gradient descent based optimization approaches to estimate the value function parameters, but fail to quantify their uncertainties. In our work, under the Kalman filtering paradigm, we establish a novel and scalable sampling framework based on stochastic gradient Markov chain Monte Carlo, which allows us to efficiently generate samples from the posterior distribution of deep neural network parameters. For TD-learning with both linear and nonlinear function approximation, we prove that the proposed algorithm converges to a stationary distribution, which allows us to measure uncertainties of the value function and its parameters.

Adversarial attacks can fool convolutional networks and make the systems vulnerable to fraud and deception. How to defend against malicious attacks is a critical challenge in practice. Adversarial attacks are often conducted by adding tiny perturbations on images to cause network misclassification. Noise reduction can defend the attacks; however, it is not suited for all the cases. Considering that different models have different tolerance abilities on adversarial attacks, we develop a novel detecting module to remove noise by adaptive process and detect adversarial attacks without modifying the models. Experimental results show that by comparing the classification results on adversarial samples of MNIST and two subclasses of ImageNet datasets, our models can successfully remove most of the noise and obtain detection accuracies of 97.71% and 92.96%, respectively. Furthermore, our adaptive module can be assembled into different networks to achieve detection accuracies of 70.83% and 71.96%, respectively, on the white-box adversarial attacks of ResNet18 and SCD01MLP images. The best accuracy of 62.5% is obtained for both networks when dealing with the black-box attacks.

Machine learning techniques in medical imaging systems are accurate, but minor perturbations in the data known as adversarial attacks can fool them. These attacks make the systems vulnerable to fraud and deception, and thus a significant challenge has been posed in practice. We present the gradient-free trained sign activation networks to detect and deter adversarial attacks on medical imaging AI systems. Experimental results show that a higher distortion value is required to attack our proposed model than the other existing state-of-the-art models on MRI, Chest X-ray, and Histopathology image datasets, where our model outperforms the best and is even twice superior. The average accuracy of our model in classifying the adversarial examples is 88.89%, whereas those for MLP and LeNet are 81.48%, and that of ResNet18 is 38.89%. It is concluded that the sign network is a solution to defend adversarial attacks due to high distortion and high accuracy on transferability. Our work is a significant step towards safe and secure medical AI systems.

In order to improve the driving safety and reduce traffic congestion during holidays and work hours, a real-time vehicle detection and counting system is a very urgently needed system. In this paper, a lane-based vehicle counting system using deep-learning networks is proposed. Our method includes YOLO vehicle detection and lane-based vehicle counting. From the vehicle detection experimental results, YOLOv3-spp has the highest Precision, Recall, and F1 score, which achieve all 100% among three YOLOv3 methods and two YOLOv2 methods. From the vehicle counting experimental results, YOLOv3-608 has the highest Accuracy, Precision and F1 scores, which achieve 91.4%, 99.3%, and 95.3% among three YOLOv3 methods, two YOLOv2 methods, and one SSD method.

Mathematical morphology is a theory and technique to collect features like geometric and topological structures in digital images. Given a target image, determining suitable morphological operations and structuring elements is a cumbersome and time-consuming task. In this paper, a morphological neural network is proposed to address this problem. Serving as a nonlinear feature extracting layer in deep learning frameworks, the efficiency of the proposed morphological layer is confirmed analytically and empirically. With a known target, a single-filter morphological layer learns the structuring element correctly, and an adaptive layer can automatically select appropriate morphological operations. For practical applications, the proposed morphological neural networks are tested on several classification datasets related to shape or geometric image features, and the experimental results have confirmed the high computational efficiency and high accuracy.

—Malignant Melanoma, known simply as Melanoma, is a type of skin cancer that appears as a mole on the skin. It is critical to detect this cancer at an early stage because it can spread across the body and may lead to the patient death. When detected early, Melanoma is curable. In this paper we propose a deep learning model (Convolutional Neural Networks) in order to automatically classify skin lesion images as Malignant or Benign. Images underwent certain pre-processing steps to diminish the effect of the normal skin region on the model. The result of the proposed model showed a significant improvement over previous work, achieving an accuracy of 97%.

Acne and Rosacea are two common skin diseases that affect many people worldwide. These two skin conditions can result in similar signs, which leads to the misdiagnosis of the case. People affected by these two skin rashes, usually tend not to seek medical diagnosis from expert dermatologists, but instead rely on over-the-counter medications and beauty products for self-treatment. Although acne and rosacea are both usually considered non-dangerous, treating acne with rosacea medication (and vice-versa) can lead to worsen symptoms. In this paper, we propose a deep learning model that can automatically distinguish Rosacea from Acne cases using infected skin images. Due to the limited number of available images, we enlarged the data set using image augmentation. Experimental results show that our model achieves a high performance with an average testing accuracy of 87.1% (over 10-folds) and 91.2% on the validation set. The good predictive performance of the model depicts its usability to classify new, unseen cases. We believe that such a model can serve as an efficient basis to build an automatic acne-rosacea distinguishing software tool.

Brain tumor classification plays an important role in brain cancer diagnosis and treatment. Pathologists typically have to work through numerous pathology images that can be in the order of hundreds or thousands which takes time and is prone to manual error. Here we investigate automating this task given pathology images as well as 3D MRI volumes without lesion maps. We use data provided by the CPM-RadPath 2019 MICCAI challenge. We first evaluate accuracy on the validation dataset with MRI and pathology images separately. We predict the 3D tumor mask with our custom developed tumor segmentation model that we used for the BraTS 2019 challenge. We show that the predicted tumor segmentations give a higher validation accuracy of 77.1% vs. 69.8% with MRI images when trained by a 3D residual convolutional neural network. For pathology images we train a 2D residual network and obtain a 66.2% validation accuracy. In both cases we find high training accuracies above 95% which suggests overfitting. We propose a dual path residual convolutional neural network model that trains simultaneously from both MRI and pathology images and we use a simple method to prevent overfitting. One path of our network is fully 3D and considers 3D tumor segmentations as input while the other path considers pathology images. To prevent overfitting we stop training after 90% training accuracy at the epoch number where our network loss increases in the following one. With this approach we achieve a validation accuracy of 84.9% showing that indeed combining the two image sources yields a better overall accuracy.

We present a high-capacity reversible, fragile, and blind watermarking scheme for medical images in this paper. A bottom-up saliency detection algorithm is applied to automatically locate the multiple arbitrarily-shaped regions of interest (ROIs). The iterative square-production algorithm is developed to generate different sizes of squares for shape decomposition on the regions of noninterest (RONIs). This scheme of combining the frequency-domain watermarking and arbitrarily-shaped ROI methods can significantly increase the watermarking capacity, whereas the embedded image fidelity is preserved. Extensive experiments were carried out on the OASIS medical image dataset, which consists of a cross-sectional collection of 416 subjects, aged from 18 to 96 years old. The results show that the proposed scheme outperforms six existing state-of-the-art schemes in terms of watermarking capacity and embedded image fidelity.

Digital image watermarking is the process of embedding and extracting watermark covertly on a carrier image. Incorporating deep learning networks with image watermarking has attracted increasing attention during recent years. However, existing deep learning-based watermarking systems cannot achieve robustness, blindness, and automated embedding and extraction simultaneously. In this paper, a fully automated image watermarking system based on deep neural networks is proposed to generalize the image watermarking processes. An unsupervised deep learning structure and a novel loss computation are proposed to achieve high capacity and high robustness without any prior knowledge of possible attacks. Furthermore, a challenging application of watermark extraction from camera-captured images is provided to validate the practicality as well as the robustness of the proposed system. Experimental results show the superiority performance of the proposed system as comparing against several currently available techniques.

Saliency detection refers to the segmentation of all visually conspicuous objects from various backgrounds. The purpose is to produce an object-mask that overlaps the salient regions annotated by human vision. In this paper, we propose an efficient bottom-up saliency detection model based on wavelet generalized lifting. It requires no kernels with implicit assumptions and prior knowledge. Multiscale wavelet analysis is performed on broadly tuned color feature channels to include a wide range of spatial-frequency information. A nonlinear wavelet filter bank is designed to emphasize the wavelet coefficients, and then a saliency map is obtained through linear combination of the enhanced wavelet coefficients. This full-resolution saliency map uniformly highlights multiple salient objects of different sizes and shapes. An object-mask is constructed by the adaptive thresholding scheme on the saliency maps. Experimental results show that the proposed model outperforms the existing state-of-the-art competitors on two benchmark datasets.

Imperceptibility, security, capacity, and robustness are among many aspects of image watermarking design. An ideal watermarking system should embed a large amount of information perfectly securely, but with no visible degradation to the host image. Many researchers have geared efforts towards developing specific techniques for variant applications. In this paper, we propose an adjustable-purpose, reversible and fragile watermarking scheme for image watermarking by particle swarm optimization (PSO). In general, given any host image and watermark, our scheme can provide an optimal watermarking solution. First, the content of a host image is analyzed to extract significant regions of interest (ROIs) automatically. The remaining regions of non-interest (RONIs) are collated for embedding watermarks by different amounts of bits determined by PSO to achieve optimal watermarking. The parameters can be adjusted relying upon user's watermarking purposes. Experimental results show that the proposed technique has accomplished higher capacity and higher PSNR (peak signal-to-noise ratio) watermarking.

A robust traffic surveillance system is crucial in improving the control and management of traffic systems. Vehicle flow processing primarily involves counting and tracking vehicles; however, due to complex situations such as brightness changes and vehicle partial occlusions, traditional image segmentation methods are unable to segment and count vehicles correctly. This paper presents a novel framework for vision-based vehicle counting and tracking, which consists of four main procedures: foreground detection, feature extraction, feature analysis, and vehicles counting/tracking. Foreground detection intends to generate regions of interest in an image, which are used to produce significant feature points. Vehicles counting and tracking are achieved by analyzing clusters of feature points. As for testing on recorded traffic videos, the proposed framework is verified to be able to separate occluded vehicles and count the number of vehicles accurately and efficiently. By comparing with other methods, we observe that the proposed framework achieves the highest occlusion segment rate and the counting accuracy.