Content uploaded by Yongsheng Gao
Author content
All content in this area was uploaded by Yongsheng Gao
Content may be subject to copyright.
A preview of the PDF is not available
Local Derivative Pattern Versus Local Binary Pattern: Face Recognition With High-Order Local Pattern Descriptor
Abstract and Figures
This paper proposes a novel high-order local pattern descriptor, local derivative pattern (LDP), for face recognition. LDP is a general framework to encode directional pattern features based on local derivative variations. The nth -order LDP is proposed to encode the ( n -1) th -order local derivative direction variations, which can capture more detailed information than the first-order local pattern used in local binary pattern (LBP). Different from LBP encoding the relationship between the central point and its neighbors, the LDP templates extract high-order local information by encoding various distinctive spatial relationships contained in a given local region. Both gray-level images and Gabor feature images are used to evaluate the comparative performances of LDP and LBP. Extensive experimental results on FERET, CAS-PEAL, CMU-PIE, Extended Yale B, and FRGC databases show that the high-order LDP consistently performs much better than LBP for both face identification and face verification under various conditions.
Figures - uploaded by Yongsheng Gao
Author content
All figure content in this area was uploaded by Yongsheng Gao
Content may be subject to copyright.
... Here, features, like local derivative pattern (LDP) [29], local gradient patterns (LGP) [30], and gray level co-occurrence matrix (GLCM) based features [31], such as contrast, energy and homogeneity are determined. Later, classification is ...
... The performance of the lung infection prediction scheme depends upon the amount of relevant information identified in the input image, and for this, it is essential to determine the prominent features. To accomplish this, the segmented image R is fed to the feature extraction process, wherein initially the image features, like LDP [29] and LGP [30] are found, and they are then are utilized for generating the GLCM features [31], like contrast, energy and homogeneity. This process is explicated in the following section. ...
A major disease affecting individuals irrespective of the different ages is lung disease and this problem is a result of different causes. The recent spread of COVID-19 caused by severe acute respiratory syndrome coronavirus 2 (SARS-CoV-2) has affected a huge community worldwide and has impacted the respiratory system adversely. The infection severity can be determined by inspecting the using X-ray images of the lung. In this work, a multilevel classification approach is presented, wherein the lung severity and COVID-19 prediction are executed based on Deep Learning (DL) technique. The contribution of this research is three-fold: (1) a novel Political Golden Search Algorithm (PGSA) was devised for routing the data accumulated from the nodes over the Internet of Things (IoT), (2) first-level classification was performed using the developed SqueezeNet-based technique, and it is optimized by the devised Taylor Political Golden Search Optimization (TPGSO) algorithm, to detect if COVID-19 is present or not, (3) second-level classification is accomplished using the TPGSO-deep convolutional neural network (DCNN) to categorize lung infection severity. The presented TPGSO-DCNN for second-level classification is examined for its performance based on testing accuracy, test negative rate (TNR), and test positive rate (TPR), and is established to have obtained values of 0.922, 0.926, and 0.909, respectively.
... The Local Binary Patterns feature extraction technique is commonly used to analyse and extract texture patterns from biometric images such as finger vein images [38], [39], [40]. LBP works by dividing an image into small local regions and analysing the pixel intensity values within each region. ...
Finger vein biometrics is an approach to identifying individuals based on the unique patterns of blood vessels in their fingers, and the technology is advanced in image capture and processing techniques, which is leading to more efficient, accurate, and reliable systems. This article focuses on a verification system that compares the matrices of an efficient finger vein verification system on different databases to test its strength and efficiency. Contrast Limited Adaptive Histogram Equalization (CLAHE) has been examined as an image enhancement and processing method to improve contrast and render details in an image easier to detect. A random forest classifier is deployed with a comparison between two pretrained systems, VGG16 and ResNet50, which are types of convolutional neural networks. VGG-16 and ResNet-50 models are implemented on three different datasets, and fine-tuning these models enabled the harnessing of their powerful capabilities and achieving superior performance on the specific image classification task.
... Riccardo et al. [42] modified the LBP-TOP descriptor using LBP and CS-LBP descriptors by combining with gradient and Gabor images, thus introducing the new descriptor CSLBP-TOP. To extract more detailed information Local Derivative Patterns (LDPs) [43] are used, which define (n−1) th order derivatives in four directions i.e. 0 0 , 45 0 , 90 0 and 135 0 for a reference pixel. An encoding function in four derivative directions is used to derive the n th order LDP from (n−1) th order derivatives. ...
Human action recognition research has become increasingly sophisticated in recent years because it is used in many different applications. Many techniques have been introduced to solve the problem of action recognition, but a number of challenges still need to be resolved. To describe a video in action recognition, it is necessary to efficiently collect and aggregate the spatial-temporal information. In this paper, we address the problem of human action recognition by combining handcrafted spatio-temporal features with deep spatial features. This paper proposed a novel method for recognizing human actions in videos by concatenating handcrafted spatio-temporal texture features extracted by our proposed feature descriptor Volume Local Derivative Gradient Ternary Patterns (VLDGTP) and deep spatial features extracted from a modified Inception-v4 network. In order to reduce the dimension and to get equal sized feature vectors, we employed the PCA dimensionality reduction technique for both types of features. Then, the dimensionality reduced feature vectors are combined and passed to the SVM classifier for action recognition. Extensive experimentation is carried out on three standard datasets, such as KTH, UCF101 and HMDB-51 datasets. The experimental results and analysis show that our proposed method (VLDGTP + DEEP_FEATURES) yields better recognition accuracy for all three benchmark datasets compared to its competitors.
... However, its performance degrades in case of intensity variations and it also contains limited information. Therefore, different variants of LBP including LCP (Local coherence pattern) [13], LTP (Local ternary pattern) [14], LDP (Local derivative pattern) [15], and LTriDP (Local tri-directional patterns) [14] had been proposed in the literature to enhance the effectiveness of the LBP. ...
Fingerprints are commonly used in biometric systems. However, the authentication of these systems became an open challenge because fingerprints can easily be fabricated. In this paper, a hybrid feature extraction approach named Integration of Probability Weighted Spatial Gradient with Ridge Orientation (IPWSGRo) has been proposed for fingerprint liveness detection. IPWSGRo integrates intensity variation and local ridge orientation information. Intensity variation is computed by using probability-weighted moments (PWM) and second order directional derivative filter. Moreover, the ridge orientation is estimated using rotation invariant Local Phase Quantization (LPQ ri) by retaining only the significant frequency components. These two feature vectors are quantized into predefined intervals to plot a 2-D histogram. The support vector machine classifier (SVM) is then used to determine the validity of fingerprints as either live or spoof. Results are obtained by applying the proposed technique on three standard databases of LivDet competition 2011, 2013, and 2015. Experimental results indicate that the proposed method is able to reduce the average classification error rates (ACER) to 5.7, 2.1, and 5.17% on LivDet2011, 2013, and 2015, respectively.
Content-Based Image Retrieval (CBIR) involves the use of algorithms designed to accurately perceive and retrieve images based on their visual content. This paper introduces the Multi-Resolution HSV-LNP Feature Fusion (MRHLFF) method which combines color and texture features. The color feature consists of three main steps starting with a Color Space Transformation (CST) step, which converts RGB images into the HSV color space, which is more suited to human color perception and robust under variable lighting conditions. The second step is HSV Histogram Computation (HHC) which involves creating normalized color histograms for Hue, Saturation, and Value channels in the HSV color space to represent color types, and brightness distribution across an image. Multi-Resolution Color Moment (MRCM) calculates six statistical moments at multiple resolutions for each HSV channel in the third step. Regarding the Texture feature, it also consists of three main steps including Grayscale Texture Preparation (GTP) which involves converting images to grayscale to focus on intensity variations in the first step. In the second step, Scale Normalization and Gradient Orientation (SNGO) standardize the size of grayscale images and analyze texture by calculating gradient orientations, generating a histogram of these orientations for consistent texture feature extraction. Finally, Rotation Invariant Texture Analysis (RITA) ensures uniform texture features by aligning images based on a dominant orientation and computing Local Neighbor Patterns (LNP) that are invariant to rotation, encapsulating detailed texture patterns into a histogram used in image analysis. Both Color and Texture features are combined into feature vectors by Feature Fusion Vector (FFV). Dimensionality Reduction via PCA (DRP) is used to enhance the data representation and retrieval efficiency. Finally, Relevance Feedback and Refinement (RF&R) uses user interactions to improve retrieval precision using the Modified k-NN Based Relevance Feedback Algorithm. This study evaluates the MRHLFF method across four standard databases: Corel 1k, Corel 10k, Brodatz, and Vistex, which shows its robustness and excellent performance in almost all cases.
Modern healthcare relies heavily on medical imaging, and as a result of its extensive use, image databases, picture archiving, and picture communication systems have been developed. Medical picture data is growing quickly due to technical advancements in the sector. These image collections provide a chance for research, teaching, and diagnosis based on evidence. As a result, suitable techniques for searching the collections for pictures with traits like the cases of interest are needed. Nevertheless, because each image in the database is enormous, processing them usually takes a lot of time. This makes retrieving medical images for diagnosis more challenging and time-consuming. Therefore, developing effective techniques for image retrieval from medical imaging collections is essential. Text that describes the scene is a major component of the traditional image retrieval techniques. They are hence keyword-based search techniques. Tracing each medical photograph by each individual scene is challenging and time-consuming. Therefore, rather than using text-based keywords to index medical photos, visual queries like colour, texture, and shape are used. Content-based medical image retrieval (CBMIR) has been a burgeoning field of study for the past 20 years. The three primary components of the CBMIR system are similarity measurement, feature extraction, and pre-processing. However, because medical images have unique properties, some of the CBMIR approaches yield disappointing results. Hence, developing effective techniques with higher retrieval accuracy and shorter retrieval times is the primary challenge of CBMIR.
Iris detection and recognition systems are commonly used by several security based applications and the present literature contains numerous solutions for the same. Due to the increased usage of iris recognition systems, several attacks are launched to compromise the security of the application. One of the most harmful attacks is the presentation attack, which aims to mess up the system or to impersonate as the legitimate user. Both these scenarios are equally serious and this work attempts to present a presentation attack detection scheme for iris images by employing deep 150 layers of Residual Network(RESNET-150) and Extreme Learning Machine (ELM) classifier. When the classifier sniffs some abnormality, then the access is denied to the malicious user. Finally, the performance of the system is tested in terms of attack detection accuracy, false positive and negative alarms.
We develop a face recognition algorithm which is insensitive to
large variation in lighting direction and facial expression. Taking a
pattern classification approach, we consider each pixel in an image as a
coordinate in a high-dimensional space. We take advantage of the
observation that the images of a particular face, under varying
illumination but fixed pose, lie in a 3D linear subspace of the high
dimensional image space-if the face is a Lambertian surface without
shadowing. However, since faces are not truly Lambertian surfaces and do
indeed produce self-shadowing, images will deviate from this linear
subspace. Rather than explicitly modeling this deviation, we linearly
project the image into a subspace in a manner which discounts those
regions of the face with large deviation. Our projection method is based
on Fisher's linear discriminant and produces well separated classes in a
low-dimensional subspace, even under severe variation in lighting and
facial expressions. The eigenface technique, another method based on
linearly projecting the image space to a low dimensional subspace, has
similar computational requirements. Yet, extensive experimental results
demonstrate that the proposed “Fisherface” method has error
rates that are lower than those of the eigenface technique for tests on
the Harvard and Yale face databases
Preface. List of Illustrations. List of Tables. 1. Introduction. 2. Face Recognition. 3. Implementation of Invariances. 4. Simple Pattern Recognition. 5. Facial Pattern Recognition. 6. Network Training. 7. Conclusions and Contributions. 8. Future Work. Index.
We have developed a near-real-time computer system that can locate and track a subject's head, and then recognize the person by comparing characteristics of the face to those of known individuals. The computational approach taken in this system is motivated by both physiology and information theory, as well as by the practical requirements of near-real-time performance and accuracy. Our approach treats the face recognition problem as an intrinsically two-dimensional (2-D) recognition problem rather than requiring recovery of three-dimensional geometry, taking advantage of the fact that faces are normally upright and thus may be described by a small set of 2-D characteristic views. The system functions by projecting face images onto a feature space that spans the significant variations among known face images. The significant features are known as "eigenfaces," because they are the eigenvectors (principal components) of the set of faces; they do not necessarily correspond to features such as eyes, ears, and noses. The projection operation characterizes an individual face by a weighted sum of the eigenface features, and so to recognize a particular face it is necessary only to compare these weights to those of known individuals. Some particular advantages of our approach are that it provides for the ability to learn and later recognize new faces in an unsupervised manner, and that it is easy to implement using a neural network architecture.