Biblio

Document scanning aims to transfer the captured photographs documents into scanned document files. However, current methods based on traditional or key point detection have the problem of low detection accuracy. In this paper, we were the first to propose a document processing system based on semantic segmentation. Our system uses OCRNet to segment documents. Then, perspective transformation and other post-processing algorithms are used to obtain well-scanned documents based on the segmentation result. Meanwhile, we optimized OCRNet's loss function and reached 97.25 MIoU on the test dataset.

In recent years, the security of AI systems has drawn increasing research attention, especially in the medical imaging realm. To develop a secure medical image analysis (MIA) system, it is a must to study possible backdoor attacks (BAs), which can embed hidden malicious behaviors into the system. However, designing a unified BA method that can be applied to various MIA systems is challenging due to the diversity of imaging modalities (e.g., X-Ray, CT, and MRI) and analysis tasks (e.g., classification, detection, and segmentation). Most existing BA methods are designed to attack natural image classification models, which apply spatial triggers to training images and inevitably corrupt the semantics of poisoned pixels, leading to the failures of attacking dense prediction models. To address this issue, we propose a novel Frequency-Injection based Backdoor Attack method (FIBA) that is capable of delivering attacks in various MIA tasks. Specifically, FIBA leverages a trigger function in the frequency domain that can inject the low-frequency information of a trigger image into the poisoned image by linearly combining the spectral amplitude of both images. Since it preserves the semantics of the poisoned image pixels, FIBA can perform attacks on both classification and dense prediction models. Experiments on three benchmarks in MIA (i.e., ISIC-2019 [4] for skin lesion classification, KiTS-19 [17] for kidney tumor segmentation, and EAD-2019 [1] for endoscopic artifact detection), validate the effectiveness of FIBA and its superiority over stateof-the-art methods in attacking MIA models and bypassing backdoor defense. Source code will be available at code.

3D reconstruction has piqued the interest of many disciplines, and many researchers have spent the last decade striving to improve on latest automated three-dimensional reconstruction systems. Three Dimensional models can be utilized to tackle a wide range of visualization problems as well as other activities. In this paper, we have implemented a method of Digital Surface Map (DSM) generation from Aerial images using Conditional Generative Adversarial Networks (c-GAN). We have used Seg-net architecture of Convolutional Neural Network (CNN) to segment the aerial images and then the U-net generator of c-GAN generates final DSM. The dataset we used is ISPRS Potsdam-Vaihingen dataset. We also review different stages if 3D reconstruction and how Deep learning is now being widely used to enhance the process of 3D data generation. We provide binary cross entropy loss function graph to demonstrate stability of GAN and CNN. The purpose of our approach is to solve problem of DSM generation using Deep learning techniques. We put forth our method against other latest methods of DSM generation such as Semi-global Matching (SGM) and infer the pros and cons of our approach. Finally, we suggest improvements in our methods that might be useful in increasing the accuracy.

Numerous prevalent techniques build a Multi-Modal Biometric (MMB) system that struggles in offering security and also revocability onto the templates. This work proffered a MMB system centred on the Modulus Fuzzy Vault (MFV) aimed at resolving these issues. The methodology proposed includes Fingerprint (FP), Palmprint (PP), Ear and also Retina images. Utilizing the Boosted Double Plateau Histogram Equalization (BDPHE) technique, all images are improved. Aimed at removing the unnecessary things as of the ear and the blood vessels are segmented as of the retina images utilizing the Modified Balanced Iterative Reducing and Clustering using Hierarchy (MBIRCH) technique. Next, the input traits features are extracted; then the essential features are chosen as of the features extracted utilizing the Bidirectional Deer Hunting optimization Algorithm (BDHOA). The features chosen are merged utilizing the Normalized Feature Level and Score Level (NFLSL) fusion. The features fused are saved securely utilizing Modulus Fuzzy Vault. Upto fusion, the procedure is repeated aimed at the query image template. Next, the de-Fuzzy Vault procedure is executed aimed at the query template, and then the key is detached by matching the query template’s and input biometric template features. The key separated is analogized with the threshold that categorizes the user as genuine or else imposter. The proposed BDPHE and also MFV techniques function efficiently than the existent techniques.

The growing internet-based services such as banking and shopping have brought both ease to human's lives and challenges in user identity authentication. Different methods have been investigated for user authentication such as retina, finger print, and face recognition. This study introduces a photoplethysmogram (PPG) based user identity authentication relying on textural features extracted from time-frequency image. The PPG signal is segmented into segments and each segment is transformed into time-frequency domain using continuous wavelet transform (CWT). Then, the textural features are extracted from the time-frequency images using Haralick's method. Finally, a classifier is employed for identity authentication purposes. The proposed system achieved an average accuracy of 99.14% and 99.9% with segment lengths of one and tweeny seconds, respectively, using random forest classifier.

Facial emotion recognition (FER) is useful in many different applications and could offer significant benefit as part of feedback systems to train children with Autism Spectrum Disorder (ASD) who struggle to recognize facial expressions and emotions. This project explores the potential of real time FER based on the use of local regions of interest combined with a machine learning approach. Histogram of Oriented Gradients (HOG) was implemented for feature extraction, along with 3 different classifiers, 2 based on k-Nearest Neighbor and 1 using Support Vector Machine (SVM) classification. Model performance was compared using accuracy of randomly selected validation sets after training on random training sets of the Oulu-CASIA database. Image classes were distributed evenly, and accuracies of up to 98.44% were observed with small variation depending on data distributions. The region selection methodology provided a compromise between accuracy and number of extracted features, and validated the hypothesis a focus on smaller informative regions performs just as well as the entire image.

A 'Completely Automated Public Turing test to tell Computers and Humans Apart' or better known as CAPTCHA is a image based test used to determine the authenticity of a user (ie. whether the user is human or not). In today's world, almost all the web services, such as online shopping sites, require users to solve CAPTCHAs that must be read and typed correctly. The challenge is that recognizing the CAPTCHAs is a relatively easy task for humans, but it is still hard to solve for computers. Ideally, a well-designed CAPTCHA should be solvable by humans at least 90% of the time, while programs using appropriate resources should succeed in less than 0.01% of the cases. In this paper, a deep neural network architecture is presented to extract text from CAPTCHA images on various platforms. The central theme of the paper is to develop an efficient & intelligent model that converts image-based CAPTCHA to text. We used convolutional neural network based architecture design instead of the traditional methods of CAPTCHA detection using image processing segmentation modules. The model consists of seven layers to efficiently correlate image features to the output character sequence. We tried a wide variety of configurations, including various loss and activation functions. We generated our own images database and the efficacy of our model was proven by the accuracy levels of 99.7%.

The subgraph matching problem arises in a number of modern machine learning applications including segmented images and meshes of 3D objects for pattern recognition, bio-chemical reactions and security applications. This graph-based problem can have a very large and complex solution space especially when the world graph has many more nodes and edges than the template. In a real use-case scenario, analysts may need to query additional information about template nodes or world nodes to reduce the problem size and the solution space. Currently, this query process is done by hand, based on the personal experience of analysts. By analogy to the well-known active learning problem in machine learning classification problems, we present a machine-based active learning problem for the subgraph match problem in which the machine suggests optimal template target nodes that would be most likely to reduce the solution space when it is otherwise overly large and complex. The humans in the loop can then include additional information about those target nodes. We present some case studies for both synthetic and real world datasets for multichannel subgraph matching.

Generalizing automated medical image segmentation methods to new image domains is inherently difficult. We have previously developed a number of automated segmentation methods that perform at the level of human readers on images acquired under similar conditions to the original training data. We are interested in exploring techniques that will improve model generalization to new imaging domains. In this study we explore a method to limit the inherent bias of these models to intensity and textural information. Using a dataset of 100 T2-weighted MR images with fat-saturation, and 100 T2-weighted MR images without fat-saturation, we explore the use of neural style transfer to induce shape preference and improve model performance on the task of segmenting the kidneys in patients affected by polycystic kidney disease. We find that using neural style transfer images improves the average dice value by 0.2. In addition, visualizing individual network kernel responses highlights a drastic difference in the optimized networks. Biasing models to invoke shape preference is a promising approach to create methods that are more closely aligned with human perception.

In recent years, the application of style transfer has become more and more widespread. Traditional deep learning-based style transfer networks often have problems such as image distortion, loss of detailed information, partial content disappearance, and transfer errors. The style transfer network based on deep learning that we propose in this article is aimed at dealing with these problems. Our method uses image edge information fusion and semantic segmentation technology to constrain the image structure before and after the migration, so that the converted image maintains structural consistency and integrity. We have verified that this method can successfully suppress image conversion distortion in most scenarios, and can generate good results.

Despite recent advances in deep learning approaches, haze mitigation in large satellite images is still a challenging problem. Due to amorphous nature of haze, object detection or image segmentation approaches are not applicable. Also it is practically infeasible to obtain ground truths for training. Bounded memory capacity of GPUs is another constraint that limits the size of image to be processed. In this paper, we propose a style transfer based neural network approach to mitigate haze in a large overhead imagery. The network is trained without paired ground truths; further, perception loss is added to restore vivid colors, enhance contrast and minimize artifacts. The paper also illustrates our use of multiple GPUs in a collective way to produce a single coherent clear image where each GPU dehazes different portions of a large hazy image.

Weakly supervised segmentation methods using bounding box annotations focus on obtaining a pixel-level mask from each box containing an object. Existing methods typically depend on a class-agnostic mask generator, which operates on the low-level information intrinsic to an image. In this work, we utilize higher-level information from the behavior of a trained object detector, by seeking the smallest areas of the image from which the object detector produces almost the same result as it does from the whole image. These areas constitute a bounding-box attribution map (BBAM), which identifies the target object in its bounding box and thus serves as pseudo ground-truth for weakly supervised semantic and instance segmentation. This approach significantly outperforms recent comparable techniques on both the PASCAL VOC and MS COCO benchmarks in weakly supervised semantic and instance segmentation. In addition, we provide a detailed analysis of our method, offering deeper insight into the behavior of the BBAM.

Semantic image annotation is a field of paramount importance in which deep learning excels. However, some application domains, like security or medicine, may need an explanation of this annotation. Explainable Artificial Intelligence is an answer to this need. In this work, an explanation is a sentence in natural language that is dedicated to human users to provide them clues about the process that leads to the decision: the labels assignment to image parts. We focus on semantic image annotation with fuzzy logic that has proven to be a useful framework that captures both image segmentation imprecision and the vagueness of human spatial knowledge and vocabulary. In this paper, we present an algorithm for textual explanation generation of the semantic annotation of image regions.

IoT is an emerging technology in modern world of communication. As the usage of IoT devices is increasing in day to day life, the secure data communication in IoT environment is the major challenge. Especially, small sized Single-Board Computers (SBCs) or Microcontrollers devices are widely used to transfer data with another in IoT. Due to the less processing power and storage capabilities, the data acquired from these devices must be transferred very securely in order to avoid some ethical issues. There are many cryptography approaches are applied to transfer data between IoT devices, but there are obvious chances to suspect encrypted messages by eavesdroppers. To add more secure data transfer, steganography mechanism is used to avoid the chances of suspicion as another layer of security. Based on the capabilities of IoT devices, low complexity images are used to hide the data with different hiding algorithms. In this research study, the secret data is encoded through QR code and embedded in low complexity cover images by applying image to image hiding fashion. The encoded image is sent to the receiving device via the network. The receiving device extracts the QR code from image using secret key then decoded the original data. The performance measure of the system is evaluated by the image quality parameters mainly Peak Signal to Noise Ratio (PSNR), Normalized Coefficient (NC) and Security with maintaining the quality of contemporary IoT system. Thus, the proposed method hides the precious information within an image using the properties of QR code and sending it without any suspicion to attacker and competes with the existing methods in terms of providing more secure communication between Microcontroller devices in IoT environment.

A synthetic aperture radar (SAR) target detection method based on the fusion of multiscale superpixel segmentations is proposed in this paper. SAR images are segmented between land and sea firstly by using superpixel technology in different scales. Secondly, image segmentation results together with the constant false alarm rate (CFAR) detection result are coalesced. Finally, target detection is realized by fusing different scale results. The effectiveness of the proposed algorithm is tested on Sentinel-1A data.

In this study, it was aimed to recognize the emotional state from facial images using the deep learning method. In the study, which was approved by the ethics committee, a custom data set was created using videos taken from 20 male and 20 female participants while simulating 7 different facial expressions (happy, sad, surprised, angry, disgusted, scared, and neutral). Firstly, obtained videos were divided into image frames, and then face images were segmented using the Haar library from image frames. The size of the custom data set obtained after the image preprocessing is more than 25 thousand images. The proposed convolutional neural network (CNN) architecture which is mimics of LeNet architecture has been trained with this custom dataset. According to the proposed CNN architecture experiment results, the training loss was found as 0.0115, the training accuracy was found as 99.62%, the validation loss was 0.0109, and the validation accuracy was 99.71%.

To ensure security, Completely Automated Public Turing test to tell Computers and Humans Apart (CAPTCHA) is widely used in people's online lives. This paper presents a Chinese character captcha sequential selection system based on convolutional neural network (CNN). Captchas composed of English and digits can already be identified with extremely high accuracy, but Chinese character captcha recognition is still challenging. The task we need to complete is to identify Chinese characters with different colors and different fonts that are not on a straight line with rotation and affine transformation on pictures with complex backgrounds, and then perform word order restoration on the identified Chinese characters. We divide the task into several sub-processes: Chinese character detection based on Faster R-CNN, Chinese character recognition and word order recovery based on N-Gram. In the Chinese character recognition sub-process, we have made outstanding contributions. We constructed a single Chinese character data set and built a 10-layer convolutional neural network. Eventually we achieved an accuracy of 98.43%, and completed the task perfectly.

The resistance to attacks aimed to break CAPTCHA challenges and the effectiveness, efficiency and satisfaction of human users in solving them called usability are the two major concerns while designing CAPTCHA schemes. User-friendliness, universality, and accessibility are related dimensions of usability, which must also be addressed adequately. With recent advances in segmentation and optical character recognition techniques, complex distortions, degradations and transformations are added to text-based CAPTCHA challenges resulting in their reduced usability. The extent of these deformations can be decreased if some additional security mechanism is incorporated in such challenges. This paper proposes an additional security mechanism that can add an extra layer of protection to any text-based CAPTCHA challenge, making it more challenging for bots and scripts that might be used to attack websites and web applications. It proposes the use of hidden text-boxes for user entry of CAPTCHA string which serves as honeypots for bots and automated scripts. The honeypot technique is used to trick bots and automated scripts into filling up input fields which legitimate human users cannot fill in. The paper reports implementation of honeypot technique and results of tests carried out over three months during which form submissions were logged for analysis. The results demonstrated great effectiveness of honeypots technique to improve security control and usability of text-based CAPTCHA challenges.

Human identification is a well-researched topic that keeps evolving. Advancement in technology has made it easy to train models or use ones that have been already created to detect several features of the human face. When it comes to identifying a human face from the side, there are many opportunities to advance the biometric identification research further. This paper investigates the human face identification based on their side profile by extracting the facial features and diagnosing the feature sets with geometric ratio expressions. These geometric ratio expressions are computed into feature vectors. The last stage involves the use of weighted means to measure similarity. This research addresses the problem of using an eXplainable Artificial Intelligence (XAI) approach. Findings from this research, based on a small data-set, conclude that the used approach offers encouraging results. Further investigation could have a significant impact on how face profiles can be identified. Performance of the proposed system is validated using metrics such as Precision, False Acceptance Rate, False Rejection Rate and True Positive Rate. Multiple simulations indicate an Equal Error Rate of 0.89.

Gait phase segmentation is the process of identifying the start and end of different phases within a gait cycle. It is essential to many medical applications, such as disease diagnosis or rehabilitation. This work utilizes inertial measurement units (IMUs) mounted on the individual's foot to gather gait information and develops a gait phase segmentation method based on the collected signals. The proposed method utilizes a weighted dynamic time warping (DTW) algorithm to measure the distance between two different gait signals, and a k-nearest neighbors (kNN) algorithm to obtain the gait phase estimates. To reduce the complexity of the DTW-based kNN search, we propose a neural network-based graph embedding scheme that is able to map the IMU signals associated with each gait cycle into a distance-preserving low-dimensional representation while also producing a prediction on the k nearest neighbors of the test signal. Experiments are conducted on self-collected IMU gait signals to demonstrate the effectiveness of the proposed scheme.

Eddy Current Pulsed Thermography (ECPT) is widely used in Nondestructive Testing (NDT) of metal defects where the defect information is sometimes affected by coil noise and edge noise, therefore, it is necessary to segment the ECPT image sequences to improve the detection effect, that is, segmenting the defect part from the background. At present, the methods widely used in ECPT are mostly based on matrix decomposition theory. In fact, tensor decomposition is a new hotspot in the field of image segmentation and has been widely used in many image segmentation scenes, but it is not a general method in ECPT. This paper analyzes the feasibility of the usage of tensor decomposition in ECPT and designs several experiments on different samples to verify the effects of two popular tensor decomposition algorithms in ECPT. This paper also compares the matrix decomposition methods and the tensor decomposition methods in terms of treatment effect, time cost, detection success rate, etc. Through the experimental results, this paper points out the advantages and disadvantages of tensor decomposition methods in ECPT and analyzes the suitable engineering application scenarios of tensor decomposition in ECPT.

Automatic Image Analysis, Image Classification, Automatic Object Recognition are some of the aspiring research areas in various fields of Engineering. Many Industrial and biological applications demand Image Analysis and Image Classification. Sample images available for classification may be complex, image data may be inadequate or component regions in the image may have poor visibility. With the available information each Digital Image Processing application has to analyze, classify and recognize the objects appropriately. Pre-processing, Image segmentation, feature extraction and classification are the most common steps to follow for Classification of Images. In this study we applied various existing edge detection methods like Robert, Sobel, Prewitt, Canny, Otsu and Laplacian of Guassian to crab images. From the conducted analysis of all edge detection operators, it is observed that Sobel, Prewitt, Robert operators are ideal for enhancement. The paper proposes Enhanced Sobel operator, Enhanced Prewitt operator and Enhanced Robert operator using morphological operations and masking. The novelty of the proposed approach is that it gives thick edges to the crab images and removes spurious edges with help of m-connectivity. Parameters which measure the accuracy of the results are employed to compare the existing edge detection operators with proposed edge detection operators. This approach shows better results than existing edge detection operators.

Advances in deep neural networks (DNNs) have shown tremendous promise in the medical domain. However, the deep learning tools that are helping the domain, can also be used against it. Given the prevalence of fraud in the healthcare domain, it is important to consider the adversarial use of DNNs in manipulating sensitive data that is crucial to patient healthcare. In this work, we present the design and implementation of a DNN-based image translation attack on biomedical imagery. More specifically, we propose Jekyll, a neural style transfer framework that takes as input a biomedical image of a patient and translates it to a new image that indicates an attacker-chosen disease condition. The potential for fraudulent claims based on such generated `fake' medical images is significant, and we demonstrate successful attacks on both X-rays and retinal fundus image modalities. We show that these attacks manage to mislead both medical professionals and algorithmic detection schemes. Lastly, we also investigate defensive measures based on machine learning to detect images generated by Jekyll.

Relevance feedback can be considered as a learning problem. It has been extensively used to improve the performance of retrieval multimedia information. In this paper, after the relevance feedback upon content-based image retrieval (CBIR) discussed, a hybrid learning scheme on multi-target retrieval (MTR) with relevance feedback was proposed. Suppose the symbolic image database (SID) of object-level with combined image metadata and feature model was constructed. During the interactive query for remote sensing image, we calculate the similarity metric so as to get the relevant image sets from the image library. For the purpose of further improvement of the precision of image retrieval, a hybrid learning scheme parameter also need to be chosen. As a result, the idea of our hybrid learning scheme contains an exception maximization algorithm (EMA) used for retrieving the most relevant images from SID and an algorithm called supported vector machine (SVM) with relevance feedback used for learning the feedback information substantially. Experimental results show that our hybrid learning scheme with relevance feedback on MTR can improve the performance and accuracy compared the basic algorithms.

In this paper, the layer choices of the image style transfer method using the VGG-19 neural network are studied. The VGG-19 network is used to extract the feature maps which have their implicit meaning as a learning basis. If the layers for stylistic learning are not suitably chosen, the quality of style transferred image may not look good. After making experiments, it can be observed that the color information is concentrated on lower layers from conv1-1 to conv2-2, and texture information is concentrated on the middle layers from conv3-1 to conv4-4. As to the higher layers from conv5-1 to conv5-4, they seem to be able to depict image content well. Based on these observations, the methods of color transfer, texture transfer and style transfer are presented and make comparisons with conventional methods.