scholarly journals Eye Localization Using Convolutional Neural Networks and Image Gradients

2018 ◽  
Author(s):  
Werton P. De Araujo ◽  
Thelmo P. De Araujo ◽  
Gustavo A. L. De Campos
Keyword(s):  
Neural Network ◽  
Neural Networks ◽  
Processing Time ◽  
Gradient Flow ◽  
Eye Detection ◽  
Regression Problem ◽  
Eye Localization ◽  
Gradient Based ◽  
Image Gradients ◽  
Facial Images

Eye detection is a preprocessing step in many methods using facial images. Some algorithms to detect eyes are based on the characteristics of the gradient flow in the iris-sclera boundary. These algorithms are usually applied to the whole face and a posterior heuristic is used to remove false positives. In this paper, we reverse that approach by using a Convolutional Neural Network (CNN) to solve a regression problem and give a coarse estimate of the eye regions, and only then do we apply the gradient-based algorithms. The CNN was combined with two gradient-based algorithms and the results were evaluated regarding their accuracy and processing time, showing the applicability of both methods for eye localization.

Weights Direct Determination of Feedforward Neural Networks without Iterative BP-Training

2010 ◽  
pp. 197-225 ◽  
Author(s):  
Yunong Zhang ◽  
Ning Tan
Keyword(s):  
Neural Network ◽  
Neural Networks ◽  
Artificial Neural Networks ◽  
Direct Determination ◽  
Bp Neural Networks ◽  
Learning Rates ◽  
The Neural Network ◽  
Gradient Based ◽  
Artificial Neural ◽  
Output Behavior

Artificial neural networks (ANN), especially with error back-propagation (BP) training algorithms, have been widely investigated and applied in various science and engineering fields. However, the BP algorithms are essentially gradient-based iterative methods, which adjust the neural-network weights to bring the network input/output behavior into a desired mapping by taking a gradient-based descent direction. This kind of iterative neural-network (NN) methods has shown some inherent weaknesses, such as, 1) the possibility of being trapped into local minima, 2) the difficulty in choosing appropriate learning rates, and 3) the inability to design the optimal or smallest NN-structure. To resolve such weaknesses of BP neural networks, we have asked ourselves a special question: Could neural-network weights be determined directly without iterative BP-training? The answer appears to be YES, which is demonstrated in this chapter with three positive but different examples. In other words, a new type of artificial neural networks with linearly-independent or orthogonal activation functions, is being presented, analyzed, simulated and verified by us, of which the neural-network weights and structure could be decided directly and more deterministically as well (in comparison with usual conventional BP neural networks).

Structured Pyramidal Neural Networks

2018 ◽  
Vol 28 (05) ◽  
pp. 1750021 ◽  
Author(s):  
Alessandra M. Soares ◽  
Bruno J. T. Fernandes ◽  
Carmelo J. A. Bastos-Filho
Keyword(s):  
Neural Network ◽  
Neural Networks ◽  
Processing Time ◽  
Receptive Fields ◽  
Memory Capacity ◽  
The Other ◽  
Support Vector ◽  
Proposed Model ◽  
Novel Method ◽  
Original Approach

The Pyramidal Neural Networks (PNN) are an example of a successful recently proposed model inspired by the human visual system and deep learning theory. PNNs are applied to computer vision and based on the concept of receptive fields. This paper proposes a variation of PNN, named here as Structured Pyramidal Neural Network (SPNN). SPNN has self-adaptive variable receptive fields, while the original PNNs rely on the same size for the fields of all neurons, which limits the model since it is not possible to put more computing resources in a particular region of the image. Another limitation of the original approach is the need to define values for a reasonable number of parameters, which can turn difficult the application of PNNs in contexts in which the user does not have experience. On the other hand, SPNN has a fewer number of parameters. Its structure is determined using a novel method with Delaunay Triangulation and k-means clustering. SPNN achieved better results than PNNs and similar performance when compared to Convolutional Neural Network (CNN) and Support Vector Machine (SVM), but using lower memory capacity and processing time.

Swarm-Based Nature-Inspired Metaheuristics for Neural Network Optimization

2018 ◽  
pp. 23-53
Author(s):  
Swathi Jamjala Narayanan ◽  
Boominathan Perumal ◽  
Jayant G. Rohra
Keyword(s):  
Neural Network ◽  
Neural Networks ◽  
Network Optimization ◽  
Network Architectures ◽  
Neural Network Optimization ◽  
Local Optima ◽  
Gradient Based ◽  
Neural Network Architectures ◽  
Nature Inspired Algorithms ◽  
Nature Inspired Metaheuristics

Nature-inspired algorithms have been productively applied to train neural network architectures. There exist other mechanisms like gradient descent, second order methods, Levenberg-Marquardt methods etc. to optimize the parameters of neural networks. Compared to gradient-based methods, nature-inspired algorithms are found to be less sensitive towards the initial weights set and also it is less likely to become trapped in local optima. Despite these benefits, some nature-inspired algorithms also suffer from stagnation when applied to neural networks. The other challenge when applying nature inspired techniques for neural networks would be in handling large dimensional and correlated weight space. Hence, there arises a need for scalable nature inspired algorithms for high dimensional neural network optimization. In this chapter, the characteristics of nature inspired techniques towards optimizing neural network architectures along with its applicability, advantages and limitations/challenges are studied.

Equation-based domain knowledge utilization into neural network structure and learning

2018 ◽  
Vol 232 (10) ◽  
pp. 1275-1291 ◽  
Author(s):  
Shrinivas Kulkarni ◽  
Anirban Guha
Keyword(s):  
Neural Network ◽  
Neural Networks ◽  
Physical Interpretation ◽  
Domain Knowledge ◽  
Feed Forward Neural Network ◽  
Regression Problem ◽  
Industrial Systems ◽  
Domain Experts ◽  
Black Boxes ◽  
Neural Network Structure

The use of neural networks as black boxes, though useful for modeling complicated industrial systems, has some limitations. No physical interpretation can be given to sections of the trained network. Incorporation of domain knowledge into neural network attempts to address this lacuna. Most of the attempts in this direction have been in the area of data classification in which sub-classes created with the help of domain experts have led to better neural networks. This work attempts to incorporate domain knowledge into the structure of a neural network for solving a regression problem—that of a piston pump leakage prediction. It shows a way in which prior knowledge about subsystems, in the form of equations, can be used to create a neural network for modeling the entire system. This approach significantly outperforms a traditional feed forward neural network. As a key contribution, this approach allows physical interpretation of the neurons which can aid in troubleshooting and anomaly detection.

scholarly journals Face Recognition with CNN and Inception Deep Learning Models

2019 ◽  
Vol 8 (3) ◽  
pp. 1932-1938
Keyword(s):  
Neural Network ◽  
Neural Networks ◽  
Deep Learning ◽  
Face Recognition ◽  
Convolutional Neural Network ◽  
Training Image ◽  
Learning Models ◽  
Training Images ◽  
Testing Data ◽  
Facial Images

In this work, deep learning methods are used to classify the facial images. ORL Database is used for the purpose of training the models and for testing. Three kinds of models are developed and their performances are measured. Convolutional Neural Networks (CNN), Convolutional Neural Network Based Inception Model with single training image per class (CNN-INC) and Convolutional Neural Network Based Inception Model with several training images per class (CNN-INC-MEAN) are developed. The ORL database has ten facial images for each person. Five images are used for training purpose and remaining 5 images are used for testing. The five images for the training are chosen randomly so that two sets of training and testing data is generated. The models are trained and tested on the two sets that are drawn from the same population. The results are presented for accuracy of face recognition

Ant Colony Optimization Applied to the Training of a High Order Neural Network with Adaptable Exponential Weights

2020 ◽  
pp. 82-95
Author(s):  
Ashraf M. Abdelbar ◽  
Islam Elnabarawy ◽  
Donald C. Wunsch II ◽  
Khalid M. Salama
Keyword(s):  
Neural Network ◽  
Neural Networks ◽  
Ant Colony Optimization ◽  
High Order ◽  
Ant Colony ◽  
Ant Colony Optimization Algorithm ◽  
Test Set ◽  
Benchmark Datasets ◽  
Gradient Based ◽  
High Order Neural Network

High order neural networks (HONN) are neural networks which employ neurons that combine their inputs non-linearly. The HONEST (High Order Network with Exponential SynapTic links) network is a HONN that uses neurons with product units and adaptable exponents. The output of a trained HONEST network can be expressed in terms of the network inputs by a polynomial-like equation. This makes the structure of the network more transparent and easier to interpret. This study adapts ACOℝ, an Ant Colony Optimization algorithm, to the training of an HONEST network. Using a collection of 10 widely-used benchmark datasets, we compare ACOℝ to the well-known gradient-based Resilient Propagation (R-Prop) algorithm, in the training of HONEST networks. We find that our adaptation of ACOℝ has better test set generalization than R-Prop, though not to a statistically significant extent.

Ant Colony Optimization Applied to the Training of a High Order Neural Network with Adaptable Exponential Weights

2016 ◽  
pp. 362-374 ◽  
Author(s):  
Ashraf M. Abdelbar ◽  
Islam Elnabarawy ◽  
Donald C. Wunsch II ◽  
Khalid M. Salama
Keyword(s):  
Neural Network ◽  
Neural Networks ◽  
Ant Colony Optimization ◽  
High Order ◽  
Ant Colony ◽  
Ant Colony Optimization Algorithm ◽  
Test Set ◽  
Benchmark Datasets ◽  
Gradient Based ◽  
High Order Neural Network

High order neural networks (HONN) are neural networks which employ neurons that combine their inputs non-linearly. The HONEST (High Order Network with Exponential SynapTic links) network is a HONN that uses neurons with product units and adaptable exponents. The output of a trained HONEST network can be expressed in terms of the network inputs by a polynomial-like equation. This makes the structure of the network more transparent and easier to interpret. This study adapts ACOR, an Ant Colony Optimization algorithm, to the training of an HONEST network. Using a collection of 10 widely-used benchmark datasets, we compare ACOR to the well-known gradient-based Resilient Propagation (R-Prop) algorithm, in the training of HONEST networks. We find that our adaptation of ACOR has better test set generalization than R-Prop, though not to a statistically significant extent.

scholarly journals Age Invariant Face Recognition using Convolutional Neural Network

2018 ◽  
Vol 8 (4) ◽  
pp. 2126 ◽  
Author(s):  
Mrudula Nimbarte ◽  
Kishor Bhoyar
Keyword(s):  
Neural Network ◽  
Neural Networks ◽  
Feature Extraction ◽  
Face Recognition ◽  
Recognition Accuracy ◽  
Classification Algorithms ◽  
The Arts ◽  
Single Structure ◽  
Facial Images ◽  
Selection Of

<span>In the recent years, face recognition across aging has become very popular and challenging task in the area of face recognition.  Many researchers have contributed in this area, but still there is a significant gap to fill in. Selection of feature extraction and classification algorithms plays an important role in this area. Deep Learning with Convolutional Neural Networks provides us a combination of feature extraction and classification in a single structure. In this paper, we have presented a novel idea of 7-Layer CNN architecture for solving the problem of aging for recognizing facial images across aging. We have done extensive experimentations to test the performance of the proposed system using two standard datasets FGNET and MORPH</span><span>(Album II). Rank-1 recognition accuracy of our proposed system is 76.6% on FGNET and 92.5% on MORPH</span><span>(Album II). Experimental results show the significant improvement over available state-of- the-arts with the proposed CNN architecture and the classifier.</span>

scholarly journals The structure of the local detector of the reprint model of the object in the image

2021 ◽  
Vol 9 (5) ◽  
pp. 7-13
Author(s):  
A. A. Kulikov
Keyword(s):  
Neural Network ◽  
Neural Networks ◽  
Visual Object ◽  
Limited Area ◽  
Visual Objects ◽  
The Face ◽  
Reduced Dimension ◽  
Canonical Version ◽  
Facial Images ◽  
Object Relative

Currently, methods for recognizing objects in images work poorly and use intellectually unsatisfactory methods. The existing identification systems and methods do not completely solve the problem of identification, namely, identification in difficult conditions: interference, lighting, various changes on the face, etc. To solve these problems, a local detector for a reprint model of an object in an image was developed and described. A transforming autocoder (TA), a model of a neural network, was developed for the local detector. This neural network model is a subspecies of the general class of neural networks of reduced dimension. The local detector is able, in addition to determining the modified object, to determine the original shape of the object as well. A special feature of TA is the representation of image sections in a compact form and the evaluation of the parameters of the affine transformation. The transforming autocoder is a heterogeneous network (HS) consisting of a set of networks of smaller dimension. These networks are called capsules. Artificial neural networks should use local capsules that perform some rather complex internal calculations on their inputs, and then encapsulate the results of these calculations in a small vector of highly informative outputs. Each capsule learns to recognize an implicitly defined visual object in a limited area of viewing conditions and deformations. It outputs both the probability that the object is present in its limited area and a set of “instance parameters” that can include the exact pose, lighting, and deformation of the visual object relative to an implicitly defined canonical version of this object. The main advantage of capsules that output instance parameters is a simple way to recognize entire objects by recognizing their parts. The capsule can learn to display the pose of its visual object in a vector that is linearly related to the “natural” representations of the pose that are used in computer graphics. There is a simple and highly selective test for whether visual objects represented by two active capsules A and B have the correct spatial relationships for activating a higher-level capsule C. The transforming autoencoder solves the problem of identifying facial images in conditions of interference (noise), changes in illumination and angle.

scholarly journals Automatic Identification of Down Syndrome Using Facial Images with Deep Convolutional Neural Network

Diagnostics ◽  
2020 ◽  
Vol 10 (7) ◽  
pp. 487 ◽  
Author(s):  
Bosheng Qin ◽  
Letian Liang ◽  
Jingchao Wu ◽  
Qiyao Quan ◽  
Zeyu Wang ◽  
...  
Keyword(s):  
Neural Network ◽  
Neural Networks ◽  
Down Syndrome ◽  
Convolutional Neural Network ◽  
Facial Recognition ◽  
Genetic Disorders ◽  
Deep Convolutional Neural Network ◽  
Automatic Identification ◽  
Deep Convolutional Neural Networks ◽  
Facial Images

Down syndrome is one of the most common genetic disorders. The distinctive facial features of Down syndrome provide an opportunity for automatic identification. Recent studies showed that facial recognition technologies have the capability to identify genetic disorders. However, there is a paucity of studies on the automatic identification of Down syndrome with facial recognition technologies, especially using deep convolutional neural networks. Here, we developed a Down syndrome identification method utilizing facial images and deep convolutional neural networks, which quantified the binary classification problem of distinguishing subjects with Down syndrome from healthy subjects based on unconstrained two-dimensional images. The network was trained in two main steps: First, we formed a general facial recognition network using a large-scale face identity database (10,562 subjects) and then trained (70%) and tested (30%) a dataset of 148 Down syndrome and 257 healthy images curated through public databases. In the final testing, the deep convolutional neural network achieved 95.87% accuracy, 93.18% recall, and 97.40% specificity in Down syndrome identification. Our findings indicate that the deep convolutional neural network has the potential to support the fast, accurate, and fully automatic identification of Down syndrome and could add considerable value to the future of precision medicine.

Sign in / Sign up

Export Citation Format

Share Document