M. Arfan Jaffar

Fully automated real time fatigue detection of drivers through Fuzzy Expert Systems

A non-intrusive fatigue detection system based on the video analysis of drivers.Eye closure duration measured through eye state information and yawning analyzed through mouth state information.Lips are searched through spatial fuzzy c-means (s-FCM) clustering.Pupils are also detected in the upper part of the face window on the basis of radii, inter-pupil distance and angle.The monitored information of eyes and mouth are further passed to Fuzzy Expert System (FES) that classifies the true state of the driver. This paper presents a non-intrusive fatigue detection system based on the video analysis of drivers. The system relies on multiple visual cues to characterize the level of alertness of the driver. The parameters used for detecting fatigue are: eye closure duration measured through eye state information and yawning analyzed through mouth state information. Initially, the face is located through Viola-Jones face detection method to ensure the presence of driver in video frame. Then, a mouth window is extracted from the face region, in which lips are searched through spatial fuzzy c-means (s-FCM) clustering. Simultaneously, the pupils are also detected in the upper part of the face window on the basis of radii, inter-pupil distance and angle. The monitored information of eyes and mouth are further passed to Fuzzy Expert System (FES) that classifies the true state of the driver. The system has been tested using real data, with different sequences recorded in day and night driving conditions, and with users belonging to different race and gender. The system yielded an average accuracy of 100% on all the videos on which it was tested.

Fuzzy anisotropic diffusion based segmentation and texture based ensemble classification of brain tumor

Abstract Brain tumor is one of the major causes of death among other types of the cancers. Proper and timely diagnosis can prevent the life of a person to some extent. Therefore we have proposed an automated reliable system for the diagnosis of the brain tumor. Proposed system is a multi-stage system for brain tumor diagnosis and tumor region extraction. First, noise removal is performed as the preprocessing step on the brain MR images. Texture features are extracted from these noise free brain MR images. Next phase of the proposed system is classification that is based on these extracted features. Ensemble based SVM classification is used. More than 99% accuracy is achieved by the classification phase. After classification, proposed system extracts tumor region from tumorous images using multi-step segmentation. First step is skull removal and brain region extraction. Next step is separating tumor region from normal brain cells using FCM clustering. Results of the proposed technique show that tumor region is extracted quite accurately. This technique has been tested against the datasets of different patients received from Holy Family hospital and Abrar MRI & CT Scan center Rawalpindi.

A modified adaptive differential evolution algorithm for color image segmentation

Image segmentation is an important low-level vision task. It is a perceptual grouping of pixels based on some similarity criteria. In this paper, a new differential evolution (DE) algorithm, modified adaptive differential evolution, is proposed for color image segmentation. The DE/current-to-pbest mutation strategy with optional external archive and opposition-based learning are used to diversify the search space and expedite the convergence process. Control parameters are automatically updated to appropriate values in order to avoid user intervention of parameters setting. To find an optimal number of clusters (the number of regions or segments), the average ratio of fuzzy overlap and fuzzy separation is used as a cluster validity index. The results demonstrate that the proposed technique outperforms state-of-the-art methods.

A dynamic fuzzy genetic algorithm for natural image segmentation using adaptive mean shift

Abstract In this paper, a colour image segmentation approach based on hybridisation of adaptive mean shift (AMS), fuzzy c-mean and genetic algorithms (GAs) is presented. Image segmentation is the perceptual faction of pixels based on some likeness measure. GA with fuzzy behaviour is adapted to maximise the fuzzy separation and minimise the global compactness among the clusters or segments in spatial fuzzy c-mean (sFCM). It adds diversity to the search process to find the global optima. A simple fusion method has been used to combine the clusters to overcome the problem of over segmentation. The results show that our technique outperforms state-of-the-art methods.AbstractIn this paper, a colour image segmentation approach based on hybridisation of adaptive mean shift (AMS), fuzzy c-mean and genetic algorithms (GAs) is presented. Image segmentation is the perceptual faction of pixels based on some likeness measure. GA with fuzzy behaviour is adapted to maximise the fuzzy separation and minimise the global compactness among the clusters or segments in spatial fuzzy c-mean (sFCM). It adds diversity to the search process to find the global optima. A simple fusion method has been used to combine the clusters to overcome the problem of over segmentation. The results show that our technique outperforms state-of-the-art methods.

Ensemble classification of pulmonary nodules using gradient intensity feature descriptor and differential evolution

For detection and classification of pulmonary nodules, there are two major issues exists in the existing computer aided diagnosis system. First major problem is automatic threshold to segment lungs and nodules. Threshold selection is a critical preprocessing step for medical images. Gaussian approximation based differential evolution has been used to find out the optimal threshold value for segmentation of lungs. Initially, 1-D histogram of the image is estimated using a blend of Gaussian functions whose parameters are calculated using the differential evolution method. Every Gaussian function estimating the histogram characterizes a pixel class and hence a threshold point. Second major problem is to extract the optimized features for classification of nodules. So, a novel gradient intensity feature descriptor for pulmonary nodule classification has been proposed using the multi-coordinate histogram of gradient and intensity based statistical features descriptor. Ensemble bagging trees has been used intelligently using the concepts of ensemble to classify the nodules. We have used standard dataset titled lung image consortium database for the verification and authentication of our proposed computer aided diagnostic (CAD) system. The proposed CAD system gives better results in comparison with existing CAD systems. The sensitivity of 97.5% is attained with an accuracy of 98.7%.

Facial expressions recognition using an ensemble of feature sets based on key-point descriptors

Abstract The authors in this study proposed facial expression recognition system in order to improve the expression recognition performance over the recently proposed systems. Feature sets for all training samples are constructed based on speed up robust feature descriptors. An ensemble of feature sets is then created incrementally. To achieve high diversity of ensemble, the dissimilarities between the training samples for each class are computed. This high diversity led to a high recognition rate. Experimentation on two publicly available datasets is performed. The system achieved 98·6% accuracy on JAFFE dataset and 96·3% accuracy on Multimedia Understanding Group dataset. The results of proposed system are compared with recently proposed work in this area and proved the soundness of the proposed method.The authors in this study proposed facial expression recognition system in order to improve the expression recognition performance over the recently proposed systems. Feature sets for all training samples are constructed based on speed up robust feature descriptors. An ensemble of feature sets is then created incrementally. To achieve high diversity of ensemble, the dissimilarities between the training samples for each class are computed. This high diversity led to a high recognition rate. Experimentation on two publicly available datasets is performed. The system achieved 98·6% accuracy on JAFFE dataset and 96·3% accuracy on Multimedia Understanding Group dataset. The results of proposed system are compared with recently proposed work in this area and proved the soundness of the proposed method.

Video scene analysis: an overview and challenges on deep learning algorithms

Video scene analysis is a recent research topic due to its vital importance in many applications such as real-time vehicle activity tracking, pedestrian detection, surveillance, and robotics. Despite its popularity, the video scene analysis is still an open challenging task and require more accurate algorithms. However, the advances in deep learning algorithms for video scene analysis have been emerged in last few years for solving the problem of real-time processing. In this paper, a review of the recent developments in deep learning and video scene analysis problems is presented. In addition, this paper also briefly describes the most recent used datasets along with their limitations. Moreover, this review provides a detailed overview of the particular challenges existed in real-time video scene analysis that has been contributed towards activity recognition, scene interpretation, and video description/captioning. Finally, the paper summarizes the future trends and challenges in video scene analysis tasks and our insights are provided to inspire further research efforts.

Brain Tumor Segmentation and Classification using Intelligent Hybrid Morphology and Diffusion

Noise present in the images degrades the image quality as well as the performance of tumor detection from images. The main objective of this research work is to improve the image quality and develop an accurate and effective automated computer-aided diagnosis system for tumor detection from brain MR images. Contourlet transform is used for image enhancement. Thresholding and morphological operators are used for detecting tumor segment. After segmentation, features extraction and classification has been performed by using Support Vector Machine and Neural Networks. The proposed method is tested on various brain MR images and this system generates good and accurate results.

An ensemble shape gradient features descriptor based nodule detection paradigm: a novel model to augment complex diagnostic decisions assistance

Primarily, there are three basic operational constituents of Nodule Detection Systems namely nodule candidate detection, classification of nodule and extraction of features. Thresholding is one of the most important factor for nodule detection. To segment the lungs and nodules, Gaussian approximation based Particle Swarm Optimization (PSO) is used to determine the optimal threshold value. After extracting lungs part, 2D and 3D region of interests (ROI’s) are used to detect nodules with area and volume information of nodules and then distinguish between wall and vessels by using fuzzy C-mean. There are three key objects namely wall, nodule and vessel in the lugs volume with specific shape. Shape-based features with Histogram of Oriented Surface Normal Vectors (HOSNV) are used as a feature descriptor. A scaled and rotation invariant multi-coordinate histogram of thegradient is used to identify nodules with different sizes and directionless shapes. So, a Novel Ensemble Shape Gradient Features (NESGF) descriptor for pulmonary nodule classification is proposed using the Histogram of Oriented Surface Normal Vectors and Multi-Coordinate Histogram of Gradient descriptor. The random forest has been used to classify the nodules through intelligent usage of the ensemble concepts to learn weak classifiers. A standard benchmark database Lung Image Consortium Database (LICD) is used for testing and validation purposes. In order to show the performance of segmentation quality, the proposed model is compared through three quantitative measures inclusive of Variation of Information (VoI), Probabilistic Rand Index (PRI) and Jaccard Measure. The methods Area Under Curve, Sensitivity, Specificity and Sensitivity have are used for classification. For classification, accuracy, sensitivity, specificity and Area under curve (AUC) has been used.

A comprehensive review of recent advances on deep vision systems

Abstract Real-time video objects detection, tracking, and recognition are challenging issues due to the real-time processing requirements of the machine learning algorithms. In recent years, video processing is performed by deep learning (DL) based techniques that achieve higher accuracy but require higher computations cost. This paper presents a recent survey of the state-of-the-art DL platforms and architectures used for deep vision systems. It highlights the contributions and challenges from over numerous research studies. In particular, this paper first describes the architecture of various DL models such as AutoEncoders, deep Boltzmann machines, convolution neural networks, recurrent neural networks and deep residual learning. Next, deep real-time video objects detection, tracking and recognition studies are highlighted to illustrate the key trends in terms of cost of computation, number of layers and the accuracy of results. Finally, the paper discusses the challenges of applying DL for real-time video processing and draw some directions for the future of DL algorithms.