Violence detection in videos using interest frame extraction and 3D convolutional neural network

Multimedia Tools and Applications - With the rapid development of detecting violent behaviors in surveillance cameras, requests on systems that automatically recognize violent events are expanded....     

A modified particle swarm optimization for economic dispatch with non-smooth cost functions

This paper presents a new approach to economic dispatch (ED) problems with non-smooth cost functions using a particle swarm optimization (PSO) technique. The practical ED problems have non-smooth cost functions with equality and inequality constraints, which makes the problem of finding the global optimum difficult when using any mathematical approaches. Since, standard PSO may converge at the early stage, in this paper, a modified PSO (MPSO) mechanism is suggested to deal with the equality and inequality constraints in the ED problems. To validate the results obtained by MPSO, standard particle swarm optimization (PSO) and guaranteed convergence particle swarm optimization (GCPSO) are applied for comparison. Also, the results obtained by MPSO, PSO and GCPSO are compared with the previous approaches reported in the literature. The results show that the MPSO produces optimal or nearly optimal solutions for the study systems.     

Concept learning by fuzzy k-NN classification and relevance feedback for efficient image retrieval

A new method for combining visual and semantic features in image retrieval is presented. A fuzzy k-NN classifier assigns initial semantic labels to database images. These labels are gradually modified by relevance feedbacks from the users. Experimental results on a database of 1000 images from 10 semantic groups are reported.     

A stochastic gravitational approach to feature based color image segmentation

In this paper, a novel image segmentation algorithm based on the theory of gravity is presented, which is called as “stochastic feature based gravitational image segmentation algorithm (SGISA)”. The proposed SGISA uses color, texture, and spatial information to partition the image into homogenous and semi-compact segments. The proposed method benefits from the advantages of both clustering and region growing image segmentation techniques. The SGISA is equipped with a new operator called “escape” that is inspired by the concept of escape velocity in physics. Moreover, motivated by heuristic search algorithms, we incorporate a stochastic characteristic with the SGISA, which gives algorithm the ability to search the image for finding the fittest regions (pixels) that are suitable for merging. Several experiments on various standard images as well as Berkley standard image database are reported. Results are compared with a well-known clustering based segmentation method, C-means, a gravitational based clustering method (SGC), and the well-known mean-shift method. The results are reported using unsupervised criteria and pre-ground-truthed measures. The obtained results confirm the effectiveness of the proposed method in color image segmentation.     

Adaptive enhancement and binarization techniques for degraded plate images

Multimedia Tools and Applications - Vehicle License Plate Recognition (VLPR) is one of the most important aspects of applying computer techniques in Intelligent Transport Systems (ITS). They face...     

Information fusion between short term learning and long term learning in content based image retrieval systems

Content based image retrieval (CBIR) systems could provide more precise results by taking the user’s feedbacks into account. Two types of the relevance feedback learning paradigms are short term learning (STL) and long term learning (LTL). By using both STL and LTL, a collaborative CBIR system is proposed in this paper. The proposed system introduced three fusion methods: including fusion in retrieved images, fusion in ranks, and fusion in similarities to make cooperation between STL and LTL. The proposed fusion methods are examined in a CBIR system equipped with a proposed statistical semantic clustering (SSC) method of LTL. The SSC method works based on the concept of semantic categories of the images by clustering techniques and constructing a relevancy matrix between images and semantic categories. The results of the SSC method with the suggested fusion methods are compared with two state-of-the-art LTL methods, namely virtual feature based method and dynamic semantic clustering. Comparative results confirm the efficiency of the proposed method. Furthermore, experimental results demonstrate that for a unique LTL method, various fusion methods lead to different results.     

A short-term learning approach based on similarity refinement in content-based image retrieval

Multimedia Tools and Applications - This paper presents a new relevance feedback approach based on similarity refinement. In the proposed approach weight correction of feature’s components is...     

A prototype classifier based on gravitational search algorithm

In recent years, heuristic algorithms have been successfully applied to solve clustering and classification problems. In this paper, gravitational search algorithm (GSA) which is one of the newest swarm based heuristic algorithms is used to provide a prototype classifier to face the classification of instances in multi-class data sets. The proposed method employs GSA as a global searcher to find the best positions of the representatives (prototypes). The proposed GSA-based classifier is used for data classification of some of the well-known benchmark sets. Its performance is compared with the artificial bee colony (ABC), the particle swarm optimization (PSO), and nine other classifiers from the literature. The experimental results of twelve data sets from UCI machine learning repository confirm that the GSA can successfully be applied as a classifier to classification problems.     

A Fast Adaptive Salt and Pepper Noise Reduction Method in Images

In this paper, an effective filtering method is proposed to remove impulse noise from images. In this two-stage method, detected noise-free pixels remain unchanged. Afterwards, a Gaussian filter with adaptive variances according to the image noise level is applied on the noisy pixels. Experimental results show that the proposed method outperforms recent impulse denoising methods in terms of PSNR, MAE, IEF, and SSIM. Moreover, the speed of the method is comparable with them, and it can be used effectively in real-time applications.     

Joint detection, channel estimation, and interference cancellation in downlink MC-CDMA communication systems using complex-valued multilayer neural networks

Data detection in the presence of interference is one of the main challenges in multicarrier code division multiple access (MC-CDMA) communication systems. In this paper, a new detection technique for downlink MC-CDMA systems is proposed. This technique uses complex-valued multilayer neural networks at the receiver side. With the new definition for desired responses (±(1+J) instead of ±1, where $ J = \sqrt {{ - 1}} $ ), the convergence rate is increased (in the training process) regarding to saturation of imaginary output and the performance is increased because of increasing Euclidean distance of output neuron inputs in two states of desired outputs (with factor of $ \sqrt {2} $ ). The performance of the proposed method is improved further by considering two various saturation coefficients (in the activation function of output layer) in the training and test processes. Since the last performance improving lead to low convergence rate, this effect is compensated by correcting the coefficient of training rate in the output layer. Simulation results confirm the high convergence rate, low computational complexity, and also good performance of the proposed method in wide range of SNRs.