GIST Descriptor Research Articles

DTDO: Driving Training Development Optimization enabled deep learning approach for brain tumour classification using MRI

ABSTRACT A brain tumour is an abnormal mass of tissue. Brain tumours vary in size, from tiny to large. Moreover, they display variations in location, shape, and size, which add complexity to their detection. The accurate delineation of tumour regions poses a challenge due to their irregular boundaries. In this research, these issues are overcome by introducing the DTDO-ZFNet for detection of brain tumour. The input Magnetic Resonance Imaging (MRI) image is fed to the pre-processing stage. Tumour areas are segmented by utilizing SegNet in which the factors of SegNet are biased using DTDO. The image augmentation is carried out using eminent techniques, such as geometric transformation and colour space transformation. Here, features such as GIST descriptor, PCA-NGIST, statistical feature and Haralick features, SLBT feature, and CNN features are extricated. Finally, the categorization of the tumour is accomplished based on ZFNet, which is trained by utilizing DTDO. The devised DTDO is a consolidation of DTBO and CDDO. The comparison of proposed DTDO-ZFNet with the existing methods, which results in highest accuracy of 0.944, a positive predictive value (PPV) of 0.936, a true positive rate (TPR) of 0.939, a negative predictive value (NPV) of 0.937, and a minimal false-negative rate (FNR) of 0.061%.

BIO-INSPIRED MULTIPLE SCALES PLACE RECOGNITION FOR ELECTRIC SUBSTATIONS

Abstract. We could get many helpful information and results from satellite remote sensing images and aerial images, including disaster monitoring, grid hidden danger identification, and electricity consumption management. In the recent years, novel computer vision and deep neural network have got a lot of attention in many fields because of mimicking mammalian cognitive mechanism as much as possible. With the in-depth of mammalian cognitive and motor mechanisms research, people trend to adopt these reliable and efficient methods for power grid management and maintenance.For utilizing computing resources and improving analysing efficiency flexibly, we propose an assessing and verification framework based on bio-inspired perception and understanding, which summarizes the most appropriate image scale in the electric facilities place recognition. The proposed framework consists of different scenes aerial images datasets, several electric facilities place recognition methods, and credible evaluating methods mimicking mammals. Firstly, we gather satellite remote images and aerial images of sufficient electric power facilities in the United States via Google Earth and other public datasets. Then, several typical place recognition methods are adopted to testing recognition ability of multi-scale perception results, like SAD, NetVLAD, and GIST descriptor. To get more reliable result, multi-units and multi-scenes experiments are implemented roundly. After all experiments and evaluations, we could get that the most appropriate image scale is 1000 m size and the highest recognition accuracy of electric power facilities location is 500 m. Conclusion in our article shows the recommended perception form and scale closest to human cognition in the power grid management and maintenance.

Vision-based Position estimation and Indoor scene recognition algorithm for Quadrotor Navigation

In this paper, an effective and simple Grid based vanishing point detection position estimation algorithm and Enhanced GIST descriptors based indoor scene recognition algorithm for navigation of MAV in indoor corridor environment is described. Two different classifiers, k-nearest neighbour classifier and support vector machine is employed for the categorization of indoor scenes into corridor, staircase or room. Indoor scene classification was performed on Dartaset-1. In the training phase of the indoor scene recognition algorithm, GIST, HODMG and Enhanced-GIST feature vectors are extracted for all the indoor training images in the Dataset-1 and indoor scene classifiers are trained for the extracted image feature vectors and assigned image labels of the indoor scenes (corridor-1, staircase-2 and room-3). In the testing phase of the indoor scene recognition algorithm, for each unknown test image frame GIST, HODMG and Enhanced-GIST feature vectors are extracted and the indoor scene classification is performed using a trained scene recognition model. The proposed indoor scene recognition algorithm using SVM with Enhanced GIST descriptors produced high recognition rates of 99.33% compared to the KNN classifiers. After recognizing the indoor scene as corridor, the MAV has to estimate its position based on the detection of vanishing point in the indoor corridor image frames. Experimental results show that the proposed method is suitable for real time operations.

Catch them alive: A malware detection approach through memory forensics, manifold learning and computer vision

The everlasting increase in usage of information systems and online services have triggered the birth of the new type of malware which are more dangerous and hard to detect. In particular, according to the recent reports, the new type of fileless malware infect the victims’ devices without a persistent trace (i.e. file) on hard drives. Moreover, existing static malware detection methods in literature often fail to detect sophisticated malware utilizing various obfuscation and encryption techniques. Our contribution in this study is two-folded. First, we present a novel approach to recognize malware by capturing the memory dump of suspicious processes which can be represented as a RGB image. In contrast to the conventional approaches followed by static and dynamic methods existing in the literature, we aimed to obtain and use memory data to reveal visual patterns that can be classified by employing computer vision and machine learning methods in a multi-class open-set recognition regime. And second, we have applied a state of art manifold learning scheme named UMAP to improve the detection of unknown malware files through binary classification. Throughout the study, we have employed our novel dataset covering 4294 samples in total, including 10 malware families along with the benign executables. Lastly, we obtained their memory dumps and converted them to RGB images by applying 3 different rendering schemes. In order to generate their signatures (i.e. feature vectors), we utilized GIST and HOG (Histogram of Gradients) descriptors as well as their combination. Moreover, the obtained signatures were classified via machine learning algorithms of j48, RBF kernel-based SMO, Random Forest, XGBoost and linear SVM. According to the results of the first phase, we have achieved prediction accuracy up to 96.39% by employing SMO algorithm on the feature vectors combined with GIST+HOG. Besides, the UMAP based manifold learning strategy has improved accuracy of the unknown malware recognition models up to 12.93%, 21.83%, 20.78% on average for Random Forest, linear SVM and XGBoost algorithms respectively. Moreover, on a commercially available standard desktop computer, the suggested approach takes only 3.56 s for analysis on average. The results show that our vision based scheme provides an effective protection mechanism against malicious applications.

Well-Known brands recognition by automated classifiers using local and global features

From color and type to patterns and illustrations, brands sense to be recognizable and convey their values and personality. Here patterns and color are key elements, as they can play a vital role in brand recognition. The images used for brand classification were handpicked and collectively named as HKDataset. We have explored various feature extractors used for classification and used automated classifiers named Linear SVM to achieve higher accuracy while tuning the model parameters to achieve optimal performance. It has been observed that Support Vector Machines performs better when using GIST descriptors combined with Bag of SIFT features. We hope to apply deep learning and other sophisticated classifiers to much-expanded categories of brands in the future.

3D Palmprint recognition using Tan and Triggs normalization technique and GIST descriptors

The 3D palmprint biometric recognition system is being considered as a very promising way to address the limitations of 2D palmprint recognition systems. This paper introduces a novel method for the 3D palmprint recognition based on the Tan and Triggs normalization technique (TT) for the preprocessing data and GIST descriptor for feature extraction. The TT technique can effectively and efficiently eliminate not only the low frequencies containing the undesirable effects of shadows but also high frequencies containing aliasing and noise in 3D palmprint. The holistic feature extraction method has been employed to attain salient characteristics. Finally, for the features matching the cosine Mahalanobis distance has been used for 3D palmprint recognition. The proposed system has been evaluated on publicly available 3D palmprint database of 8000 samples. Experimental analyses show that the proposed method can functionally eradicate the effect of uneven illumination and greatly improve the performance of the recognition system. Moreover, experimental results demonstrate that our method is capable of competing with many existing state-of-the-art 3D palmprint recognition techniques as well as outperforming many others.

Ensemble Learning Prediction of Drug-Target Interactions Using GIST Descriptor Extracted from PSSM-Based Evolutionary Information.

Identifying the drug-target interactions (DTIs) plays an essential role in new drug development. However, there still has the limited knowledge of DTIs and a significant number of unknown DTI pairs. Moreover, the traditional experimental methods have inevitable disadvantages such as high cost and time-consuming. Therefore, developing computational methods for predicting DTIs is attracting more and more attention. In this study, we report a novel computational approach for predicting DTI using GIST feature, position-specific scoring matrix (PSSM), and rotation forest (RF). Specifically, each target protein is first converted into a PSSM for retaining evolutionary information. Then, the GIST feature is extracted from PSSM and substructure fingerprint information is adopted to extract the feature of the drug. Finally, combining each protein and drug features to form a new drug-target pair, which is employed as input feature for RF classifier. In the experiment, the proposed method achieves high average accuracies of 89.25%, 85.93%, 82.36%, and 73.89% on enzyme, ion channel, G protein-coupled receptors (GPCRs), and nuclear receptor, respectively. For further evaluating the prediction performance of the proposed method, we compare it with the state-of-the-art support vector machine (SVM) classifier on the same golden standard dataset. These promising results illustrate that the proposed method is more effective and stable than other methods. We expect the proposed method to be a useful tool for predicting large-scale DTIs.

A survey on automatic image annotation

Automatic image annotation is a crucial area in computer vision, which plays a significant role in image retrieval, image description, and so on. Along with the internet technique developing, there are numerous images posted on the web, resulting in the fact that it is a challenge to annotate images only by humans. Hence, many computer vision researchers are interested in automatic image annotation and make a great effort in optimizing its performance. Automatic image annotation is a task that assigns several tags in a limited vocabulary to describe an image. There are many algorithms proposed to tackle this problem and all achieve great performance. In this paper, we review seven algorithms for automatic image annotation and evaluate these algorithms leveraging different image features, such as color histograms and Gist descriptor. Our goal is to provide insights into the automatic image annotation. A lot of comprehensive experiments, which are based on Corel5K, IAPR TC-12, and ESP Game datasets, are designed to compare the performance of these algorithms. We also compare the performance of traditional algorithms employing deep learning features. Considering that not all associated labels are annotated by human annotators, we leverage the DIA metrics on IAPR TC-12 and ESP Game datasets.

Brand Recognition of Phishing Web Pages via Global Image Descriptors

Phishing attacks, which have exponentially increased in recent years, are a form of cyber attack aiming to steal sensitive credentials of innocent users. In general, the attackers attempt to deceive users by creating and submitting a fake but visually similar version of a legitimate web page, which has already been in usage. In this study, we suggest an approach for recognition of phishing web pages by utilizing two global image descriptors namely GIST and local binary patterns (LBP) which have never been employed in phishing web page recognition literature. Moreover, in order to obtain a discriminative representation, we have experimented two kinds of visual feature extraction scheme such as (1) “holistic” and (2) “multi-level patches”. While we have only used whole web page screenshot in “holistic” scheme, screenshots were divided into equally sized smaller crops at growing number of levels during the implementation of “multi-level” patches scheme. In order to evaluate the proposed approach, we have employed a publicly available phishing web page dataset in literature including screenshots of both 14 different highly phished brands and legitimate web pages posing an open-set problem for researchers. Besides, the aforementioned dataset covers 1313 training and 1539 testing cases in total. The visual signatures extracted by use of GIST and LBP descriptors were then fed to various machine learning models such as SVM, Random Forest and XGBoost (regularized gradient tree boosting). According to the results of comprehensively conducted experiments, XGBoost has been found as the best learner. In line with this finding, we obtained 87.7% (GIST) and 83.1% (LBP) validation accuracy along with the representation of “multi-level patches”. Consequently, it has been shown that preferred global image descriptors can be successfully employed for detecting and recognizing phishing web pages. In addition, average required time for processing one screenshot (around 1.12 sec.) with GIST descriptors indicates that the proposed scheme and GIST can be effectively used as a browser based plug-in for recognizing brands of phishing web pages.

Automated detection of fault types in CCTV sewer surveys

Abstract Sewers must be regularly inspected to prioritise effective maintenance, which can be an expensive and time-consuming process. This paper presents a methodology to automatically identify the type of a detected fault using raw closed circuit television (CCTV) footage. The procedure calculates the GIST descriptor of a video frame containing a fault before applying a collection of random forest classifiers to identify the fault's type. Order oblivious filtering is used to further improve the methodology's performance on continuous footage. The technology, including various classifier architectures, has been validated and demonstrated on CCTV footage collected by Wessex Water. The methodology achieved a peak accuracy of 73% when applied to well-represented fault types, showing promise for future application in the water industry.

An Improved Multispectral Palmprint Recognition System Using Autoencoder with Regularized Extreme Learning Machine.

Multispectral palmprint recognition system (MPRS) is an essential technology for effective human identification and verification tasks. To improve the accuracy and performance of MPRS, a novel approach based on autoencoder (AE) and regularized extreme learning machine (RELM) is proposed in this paper. The proposed approach is intended to make the recognition faster by reducing the number of palmprint features without degrading the accuracy of classifier. To achieve this objective, first, the region of interest (ROI) from palmprint images is extracted by David Zhang's method. Second, an efficient normalized Gist (NGist) descriptor is used for palmprint feature extraction. Then, the dimensionality of extracted features is reduced using optimized AE. Finally, the reduced features are fed to the RELM for classification. A comprehensive set of experiments are conducted on the benchmark MS-PolyU dataset. The results were significantly high compared to the state-of-the-art approaches, and the robustness and efficiency of the proposed approach are revealed.

Revisiting Gist-PCA Hashing for Near Duplicate Image Detection

This paper presents a scalable method of near duplicate image detection based on Gist-PCA (principal component analysis) hashing. While most of transform coding methods have been interested in nearest neighbor search with applications to similar image search, we solve a range search problems found in near duplicate detection problems. At first, we argue that the PCA hashing of the Gist descriptor is adequate for near duplicate image detection. Then, we decompose the Gist-PCA binary code into a hash key and a residual binary code for scalability into large-scale datasets. In addition, a multi-block approach is incorporated into the method to deal with strong variations, such as image cropping and border framing. Experimental results show that the proposed method is more accurate and faster than the real-valued Gist descriptor and other nearest neighbor search methods.

Automatic texture exemplar extraction based on global and local textureness measures

Texture synthesis is widely used for modeling the appearance of virtual objects. However, traditional texture synthesis techniques emphasize creation of optimal target textures, and pay insufficient attention to choice of suitable input texture exemplars. Currently, obtaining texture exemplars from natural images is a labor intensive task for the artists, requiring careful photography and significant postprocessing. In this paper, we present an automatic texture exemplar extraction method based on global and local textureness measures. To improve the efficiency of dominant texture identification, we first perform Poisson disk sampling to randomly and uniformly crop patches from a natural image. For global textureness assessment, we use a GIST descriptor to distinguish textured patches from non-textured patches, in conjunction with SVM prediction. To identify real texture exemplars consisting solely of the dominant texture, we further measure the local textureness of a patch by extracting and matching the local structure (using binary Gabor pattern (BGP)) and dominant color features (using color histograms) between a patch and its sub-regions. Finally, we obtain optimal texture exemplars by scoring and ranking extracted patches using these global and local textureness measures. We evaluate our method on a variety of images with different kinds of textures. A convincing visual comparison with textures manually selected by an artist and a statistical study demonstrate its effectiveness.

Indoor Scene Recognition for Micro Aerial Vehicles Navigation using Enhanced-GIST Descriptors

<p>An indoor scene recognition algorithm combining histogram of horizontal and vertical directional morphological gradient features and GIST features is proposed in this paper. New visual descriptor is called enhanced-GIST. Three different classifiers, k-nearest neighbour classifier, Naïve Bayes classifier and support vector machine, are employed for the classification of indoor scenes into corridor, staircase or room. The evaluation was performed on two indoor scene datasets. The scene recognition algorithm consists of training phase and a testing phase. In the training phase, GIST, CENTRIST, LBP, HODMG and enhanced-GIST feature vectors are extracted for all the training images in the datasets and classifiers are trained for these image feature vectors and image labels (corridor-1, staircase-2 and room-3). In the test phase, GIST, CENTRIST, LBP, HODMG and enhanced-GIST feature vectors are extracted for each unknown test image sample and classification is performed using a trained scene recognition model. The experimental results show that indoor scene recognition algorithm employing SVM with enhanced GIST descriptors produces very high recognition rates of 97.22 per cent and 99.33 per cent for dataset-1 and dataset-2, compared to kNN and Naïve Bayes classifiers. In addition to its accuracy and robustness, the algorithm is suitable for real-time operations.</p>

Discriminative Deep Belief Network for Indoor Environment Classification Using Global Visual Features

Indoor environment classification, also known as indoor environment recognition, is a highly appreciated perceptual ability in mobile robots. In this paper, we present a novel approach which is centered on biologically inspired methods for recognition and representation of indoor environments. First, global visual features are extracted by using the GIST descriptor, and then we use the subsequent features for training the discriminative deep belief network (DDBN) classifier. DDBN employs a new deep architecture which is based on restricted Boltzmann machines (RBMs) and the joint density model. The back-propagation technique is used over the entire classifier to fine-tune the weights for an optimum classification. The acquired experimental results validate our approach as it performs well both in the real-world and in synthetic datasets and outperforms the Convolution Neural Networks (ConvNets) in terms of computational efficiency.

A novel incremental topological mapping using global visual features

Mapping is fundamental in the navigation task of autonomous mobile robots. In appearance-based mapping, the process of detecting visual loop closing determines whether the current observation comes from a previously visited location or a new one. The purpose of this paper is to present a new method of exploring indoor environments by an autonomous mobile robot, as well as building topological maps based on global visual attributes. This method takes advantage of the small size of the GIST descriptors, and the ease of their calculation. We also make use of omnidirectional images to build a single global visual descriptor showing an entire room. Furthermore, in order to handle the problem of a visual loop closing, we have employed a formula that correctly assigns each global descriptor to its location.

Multi-modal framework for automatic detection of diagnostically important regions in nonalcoholic fatty liver ultrasonic images

The severity of fat in ultrasonic liver images is quantified based on characteristics of three regions in the image namely diaphragm, periportal veins and texture of liver parenchyma. The characteristics of these regions vary with the severity of fat in the liver, and is subjected to low signal to noise ratio, low contrast, poorly defined organ boundaries, etc., hence locating these regions in ultrasound images is challenging task for the sonographers. Automated detection of these regions will help the sonographers to do accurate diagnosis in shorter time, and also acts as a fundamental step to develop automated diagnostic algorithms. In this paper, we propose a novel multi-modal framework for detecting diaphragm, periportal veins and texture of liver parenchyma in ultrasonic liver ultrasound images. Since the characteristics of these regions differ from each other, we propose a specific algorithm for detecting each region. Diaphragm and periportal veins are detected with the combination of Viola Jones and GIST descriptor based classifier, while homogeneous texture regions are detected with the combination of histogram features based classifier and connected components algorithm. The proposed algorithm when tested on 180 ultrasound liver images, detected the diaphragm, periportal veins and texture regions with an accuracy of 97%, 91% and 100% respectively.

Novel expert system for glaucoma identification using non-parametric spatial envelope energy spectrum with fundus images

Glaucoma is the prime cause of blindness and early detection of it may prevent patients from vision loss. An expert system plays a vital role in glaucoma screening, which assist the ophthalmologists to make accurate decision. This paper proposes a novel technique for glaucoma detection using optic disk localization and non-parametric GIST descriptor. The method proposes a novel area based optic disk segmentation followed by the Radon transformation (RT). The change in the illumination levels of Radon transformed image are compensated using modified census transformation (MCT). The MCT images are then subjected to GIST descriptor to extract the spatial envelope energy spectrum. The obtained dimension of the GIST descriptor is reduced using locality sensitive discriminant analysis (LSDA) followed by various feature selection and ranking schemes. The ranked features are used to build an efficient classifier to detect glaucoma. Our system yielded a maximum accuracy (97.00%), sensitivity (97.80%) and specificity (95.80%) using support vector machine (SVM) classifier with nineteen features. Developed expert system also achieved maximum accuracy (93.62%), sensitivity (87.50%) and specificity (98.43%) for public dataset using twenty six features. The proposed method is efficient and computationally less expensive as it require only nineteen features to model a classifier for the huge dataset. Therefore the proposed method can be effectively utilized in hospitals for glaucoma screening.

Visual Localization across Seasons Using Sequence Matching Based on Multi-Feature Combination.

Visual localization is widely used in autonomous navigation system and Advanced Driver Assistance Systems (ADAS). However, visual-based localization in seasonal changing situations is one of the most challenging topics in computer vision and the intelligent vehicle community. The difficulty of this task is related to the strong appearance changes that occur in scenes due to weather or season changes. In this paper, a place recognition based visual localization method is proposed, which realizes the localization by identifying previously visited places using the sequence matching method. It operates by matching query image sequences to an image database acquired previously (video acquired during traveling period). In this method, in order to improve matching accuracy, multi-feature is constructed by combining a global GIST descriptor and local binary feature CSLBP (Center-symmetric local binary patterns) to represent image sequence. Then, similarity measurement according to Chi-square distance is used for effective sequences matching. For experimental evaluation, the relationship between image sequence length and sequences matching performance is studied. To show its effectiveness, the proposed method is tested and evaluated in four seasons outdoor environments. The results have shown improved precision–recall performance against the state-of-the-art SeqSLAM algorithm.

An efficient traffic sign recognition based on graph embedding features

Traffic sign recognition (TSR) is one of the significant modules of an intelligent transportation system. It instantly assists the drivers to efficiently recognize the traffic sign. Recognition of traffic sign is a large-scale feature learning problem with different real-world appearances. The main goal of this paper is to develop an efficient TSR method, which can run on an ordinary personal computer (PC). In the proposed method, GIST descriptors of the traffic sign images are extracted and subjected to graph-based linear discriminant analysis to reduce the dimension. Moreover, it effectively learns the discriminative subspace through the graph structure with increased computational efficiency. An efficient TSR module is built by conducting series of experiments using support vector machine, extreme learning machine, and k-nearest neighbor (k-NN) classifiers on available public datasets. Our approach achieved the highest recognition accuracy of 96.33 and 97.79% using k-NN classifier for German Traffic Sign Recognition Benchmark (GTSRB) and Belgium Traffic Sign Classification Benchmark (BelgiumTSC), respectively. Also it achieved 99.1% accuracy for a subcategory of GTSRB traffic signs and able to predict the class of unknown traffic sign within 0.0019 s on an ordinary PC. Hence, it can be used in real-world driver assistance system.