Showing papers by "Adnan Khashman published in 2017"

Deep learning in vision-based static hand gesture recognition

Abstract: Hand gesture for communication has proven effective for humans, and active research is ongoing in replicating the same success in computer vision systems. Human–computer interaction can be significantly improved from advances in systems that are capable of recognizing different hand gestures. In contrast to many earlier works, which consider the recognition of significantly differentiable hand gestures, and therefore often selecting a few gestures from the American Sign Language (ASL) for recognition, we propose applying deep learning to the problem of hand gesture recognition for the whole 24 hand gestures obtained from the Thomas Moeslund’s gesture recognition database. We show that more biologically inspired and deep neural networks such as convolutional neural network and stacked denoising autoencoder are capable of learning the complex hand gesture classification task with lower error rates. The considered networks are trained and tested on data obtained from the above-mentioned public database; results comparison is then made against earlier works in which only small subsets of the ASL hand gestures are considered for recognition.

Non-Destructive Prediction of Concrete Compressive Strength Using Neural Networks

Abstract: Our thirst for progress as humans is reflected by our continuous research activities in different areas leading to many useful emerging applications and technologies. Artificial intelligence and its applications are good examples of such explored fields with varying expectations and realistic results. Generally, artificially intelligent systems have shown their capability in solving real-life problems; particularly in non-linear tasks. Such tasks are often assigned to an artificial neural network (ANN) model to arbitrate as they mimic the structure and function of a biological brain; albeit at a basic level. In this paper, we investigate a newly emerging application area for ANNs; namely civil engineering. We design, implement and test an ANN model to predict and classify the compressive strength of different concrete mixes into low, moderate or high strength. Traditionally, the performance of concrete is affected by many non-linear factors and testing its strength comprises a destructive procedure of concrete samples. Numerical results in this work show high efficiency in correctly classifying the compressive strength, thus making it possible to use in real-life applications.

Automatic system for grading banana using GLCM texture feature extraction and neural network arbitrations

Abstract: Nowadays, an identification system is needed in the food processing industries to boost the efficiency of production so as to meet up with demand in the society. Manual approach is often used in product grading and quality control, and this unfortunately could lead to uneven products, higher time expense, and fatigue by the human operators. Therefore, we propose in this article, an automatic system for classification of banana whether it is healthy for production or not. Such a system is faster, accurate and also relieves the stress that an operator may have. Our system uses GLCM texture feature analysis to extract the features required for training and testing three classification models; namely, radial basis function (RBF), support vector machine (SVM), and backpropagation neural network (ANN). A classification performance comparison is drawn between the different classification models, and the obtained experimental results indicate that such intelligent grading systems may be efficiently used in real life applications for similar tasks in food processing industries. Practical applications The Automatic system is highly needed in food processing industries to meet up with the production of food products required in the societies. Healthy food products are needed in the society and this can be achieved by implementing a system that will enhance in the sorting or grading of the raw materials (such as banana) used in food processing industries. This system is accurate, economical, and faster in achieving the best product. Such a system will make the product be readily available in the market i.e. meeting the need of the people.

Iris nevus diagnosis: convolutional neural network and deep belief network

Abstract: This work presents the diagnosis of iris nevus using a convolutional neural network (CNN) and deep belief network (DBN). Iris nevus is a pigmented growth (tumor) found in the front of the eye or around the pupil. It is seen that racial and environmental factors affect the iris color (e.g., blue, hazel, brown) of patients; hence, pigmented growths may be masked in the eye background or iris. In this work, some image processing techniques are applied to images to reinforce areas of interests in them, after which the considered classifiers are trained. We describe the automated diagnosis of iris nevus using neural network-based systems for the classification of eye images as “nevus affected” and “unaffected”. Recognition rates of 93.35% and 93.67% were achieved for the CNN and DBN, respectively. Hence, the systems described in this work can be used satisfactorily for diagnosis or to reinforce the confidence in manual-visual diagnosis by medical experts.

Banknote recognition: investigating processing and cognition framework using competitive neural network.

Abstract: Humans are apt at recognizing patterns and discovering even abstract features which are sometimes embedded therein. Our ability to use the banknotes in circulation for business transactions lies in the effortlessness with which we can recognize the different banknote denominations after seeing them over a period of time. More significant is that we can usually recognize these banknote denominations irrespective of what parts of the banknotes are exposed to us visually. Furthermore, our recognition ability is largely unaffected even when these banknotes are partially occluded. In a similar analogy, the robustness of intelligent systems to perform the task of banknote recognition should not collapse under some minimum level of partial occlusion. Artificial neural networks are intelligent systems which from inception have taken many important cues related to structure and learning rules from the human nervous/cognition processing system. Likewise, it has been shown that advances in artificial neural network simulations can help us understand the human nervous/cognition system even furthermore. In this paper, we investigate three cognition hypothetical frameworks to vision-based recognition of banknote denominations using competitive neural networks. In order to make the task more challenging and stress-test the investigated hypotheses, we also consider the recognition of occluded banknotes. The implemented hypothetical systems are tasked to perform fast recognition of banknotes with up to 75 % occlusion. The investigated hypothetical systems are trained on Nigeria’s Naira banknotes and several experiments are performed to demonstrate the findings presented within this work.

A simple and practical review of over-fitting in neural network learning

Abstract: Training a neural network involves the adaptation of its internal parameters for modelling a specific task. The states of the internal parameters during training describe how much experiential knowledge the model has acquired. Although, it is desirable that a trained neural network achieves zero classification error on the training examples while tuning its internal parameters for a task, the amount of generalisation power that is lost while enforcing such a learning constraint on the model is quite important. In this paper, we review from a practical perspective the consequences of enforcing such a learning constraint which results in a model that has learned a smooth mapping function or essentially 'memorised' the training data. In addition, we investigate how the curse of dimensionality relates to such a learning constraint. For our experiments, we consider handwritten character recognition applications using publicly available datasets.

Performance Evaluation of Binarization Methods for Document Images

Abstract: In scanned documents, where noise, contrast, and illumination vary, classifying pixels as foreground or background pixels is still a difficult and challenging problem. Several evaluation studies on binarization methods for document images were previously performed, however, performing an objective evaluation to determine an optimal binarization method is not trivial because of the application-dependency of the different methods and the varieties in document databases. In this paper, the aim is to determine an optimal binarization method that can be effectively used with a variety of scanned documents. Firstly, a comparative study of thirteen binarization methods applied to gray level images of degraded historical documents, artificially created words, and handwritten documents is presented. Secondly, three new image quality parameters, used for performance evaluation in addition to visual inspection of binarized images are proposed. Experimental results suggest that local method Water Flow Model and global methods Kapur and Otsu methods outperform the other ten binarization methods on all images.