The massive volume of images produced in recent years has made image retrieval one of the topics of research in the field of Machine vision and image processing. The main challenge of content-based image retrieval systems is to extract the appropriate feature vector for image description to enable image retrieval effectively. In this research, a content-based image retrieval framework is introduced. The introduced feature vector is a combination of low-level features and mid-level features of the image. Extraction of low-level features of the image, including color, shape and texture, was performed using multi-level autocorrelation, discrete wavelet transform and fractal dimension analysis. Mid-level features are also extracted using the Deep Boltzmann Machine and by learning the low-level features of the image. The resulting feature vector is adjusted with 1K Corel database images and the performance of the proposed framework is also measured on 5K and 10K Corel databases. The best evaluation results are reported on 99.5%, 99.2% and 99.6% of the mentioned databases, respectively.

The massive volume of images produced in recent years has made image retrieval one of the topics of research in the field of Machine vision and image processing. The main challenge of content-based image retrieval systems is to extract the appropriate feature vector for image description to enable image retrieval effectively. In this research, a content-based image retrieval framework is introduced. The introduced feature vector is a combination of low-level features and mid-level features of the image. Extraction of low-level features of the image, including color, shape and texture, was performed using multi-level autocorrelation, discrete wavelet transform and fractal dimension analysis. Mid-level features are also extracted using the Deep Boltzmann Machine and by learning the low-level features of the image. The resulting feature vector is adjusted with 1K Corel database images and the performance of the proposed framework is also measured on 5K and 10K Corel databases. The best evaluation results are reported on 99. 5%, 99. 2% and 99. 6% of the mentioned databases, respectively.

Nowadays, ground vehicle monitoring (GVM) is one of the areas of application in the intelligent traffic control system using image processing methods. In this context, the use of unmanned aerial vehicles based on thermal infrared (UAV-TIR) images is one of the optimal options for GVM due to the suitable spatial resolution, costeffective and low volume of images. The methods that have been proposed for vehicle extraction from thermal infrared imaging often experience problems such as low accuracy in detection, segmentation (e. g. HOG+SVM) and also the need for big data training (e. g. Deep learning methods). In the present study, a new model, called SegRBM-Net, based on Deep learning (DL) and the restricted Boltzmann Machine (RBM) is being presented. One of the features of the SegRBM-Net model is the improving accuracy of vehicle detection and segmentation from thermal infrared images by using both convolutional layers and the features of the Gaussian-Bernoulli restricted Boltzmann Machine. This structure has led the algorithm to find the target faster and more accurately than other DL methods. To examine the performance of the proposed method, we performed a controlled benchmark (e. g. high density of vehicles scene, and difference in viewing angle) of SegRBM-Net and other DL models on four UAV-TIR image datasets. The results showed that the SegRBM-Net model with a mean accuracy of 99% and improved processing speed compared with similar methods have a good performance.

In this paper, we show how to obtain suitable differential characteristics for block ciphers with neural networks. We represent the operations of a block cipher, regarding their differential characteristics, through a directed weighted graph. In this way, the problem of finding the best differential characteristic for a block cipher reduces to the problem of finding the minimum-weight multi-path way between two known nodes in the proposed graph. We applied Hopfield network to find the minimum-weight multi-path way. In this technique, the probability of convergence to a local minimum increases when the number of rounds of the cipher increases. We also applied Boltzmann Machine to avoid local minima. We applied these techniques to find 3-round, 4-round and 5-round differential characteristics of Serpent block cipher, and repeated the optimization procedures for each characteristics 100 times. With Hopfield network, we obtained suitable results 100, 20 and 1 times for 3-round, 4-round and 5-round of the Serpent respectively. With Boltzmann Machine, we obtained suitable results 100, 99 and 30 times for 3-round, 4- round and 5-round of the Serpent respectively. These results show that simulated annealing help avoiding the many local minima of energy function. We compare the probabilities of our obtained differential characteristics for Serpent with the probabilities of eight differential characteristics previously reported in other papers. The comparison shows that our proposed technique obtains better results in 6 cases, and the same results in 2 cases. We also found a 7-round differential characteristic with a probability of 2-125 with Boltzmann Machine. Neglecting the reported Bommerang differential characteristics of Serpent, our obtained 7-round differential characteristic is the first report on a differential characteristic for more than 6 rounds of this cipher. The results of experiments indicate the efficiency of neural networks to find suitable differential characteristics of block ciphers.    

Recently, a number of Extreme Learning Machine (ELM) based training algorithms have been introduced for training Deep neural network structures. ELM based Auto-Encoder (ELM-AE) is one such algorithm that has been used for making multilayer structures and tuning parameters of each layer. In a simple ELM-AE training algorithm, the weights of the first layer are initialized randomly. This issue is a leading factor in producing reconstruction error. The frequent use of ELM-AE in Deep network layers results in propagating such errors through Deep structures and in decreasing performance as a consequent. In this paper, we introduce a multilayer structure and a new learning algorithm to train it that prevents error propagation. In order to boost the performance of the model, the parameters in the first layer are initialized by a novel type of ELM-AE called Repeated-AE (RAE) rather than by a random selection method. This RAE-based technique determines the parameters in the first layer far better than do the other ELM-AE existed methods. Next, a single hidden layer ELM is applied for handling the classification task. Experimental results for data classification show that the proposed method outperforms some other methods in terms of the average accuracy over all datasets by amounts of 4%, 26%, 17% and 31%. Eventually, so as to verify the performance of the proposed multilayer ELM-AE in application, we used this model to reconstruct images. The reconstructed images obtained by our approach appeared visually a lot better compared to those obtained by the other methods do.