No abstract

A B S T R A C TNowadays, with huge progress in digital imaging, new image processing methods are needed to manage digital images stored on disks. Image retrieval has been one of the most challengeable fields in digital image processing which means searching in a big database in order to represent similar images to the query image. Although many efficient researches have been performed for this topic so far, there is a semantic gap between human concept and features extracted from the images and it has become an important problem which decreases retrieval precision. In this paper, a convolutional neural network (CNN) is used to extract deep and high-level features from the images. Next, an optimization problem is defined in order to model the retrieval system. Heuristic algorithms such as genetic algorithm (GA) and particle swarm optimization (PSO) have shown an effective role in solving the complex problems. A recent introduced heuristic algorithm is Grasshopper Optimization Algorithm (GOA) which has been proved to be able to solve difficult optimization problems. So, a new search method, modified grasshopper optimization algorithm (MGOA) is proposed to solve modeled problem and to retrieve similar images efficiently, despite of total search in database. Experimental results showed that the proposed system named CNN-MGOA achieves superior accuracy compared to traditional methods.

Detecting same people in different surveillance cameras, named person re-identification, has become a challenging and critical task in image processing. Since surveillance images usually have low resolution and different viewpoints, matching persons on them is still difficult. In this paper, a proposed method for person re-identification is introduced based on exploring similarity in different depth layers of convolutional neural network (CNN). To this end, after determining each person as a category for training CNN, optimum filters are obtained to find the best discriminative feature maps based on them. Smoothed discriminative features (SDF) are defined to compute similarity between persons. Experimental results, performed on CUHK01 database, demonstrate that the proposed method outperforms state-of-the-art feature extraction methods for person re-identification.

Automatic waveform recognition has become an important task in radar systems and spread spectrum communications. Identifying the modulation of received signals helps to recognize different invader transmitters. In this paper, a noise aware model is proposed to recognize the modulation type based on time-frequency characteristics. To this end, Choi-Williams representation is used to obtain spatial 2D pattern of received signal. After that, a deep model is constructed to make signal clear from noise and extract robust and discriminative features from time-frequency pattern, based on auto-encoder and Convolutional Neural Networks (CNN). In order to reduce the effect of noise and adversarial disorders, a new database of different modulation patterns with different AWGN noises and fading Rayleigh channel is created which helps model to avoid the effects of noise on modulation recognition. Our database contains radar modulations such as Barker, LFM, Costas and Frank code which are known as frequently used modulations on wireless communication. Infact, the main novelty of this work is designing this database and proposing noise-aware model. Experimental results demonstrate that the proposed model achieves superior performance for automatic classification recognition with 99.24% of accuracy in noisy medium with minimum SNR of -5dB while the accuracy is 97.90% in SNR of -5dB and f=15 Hz of Doppler frequency. Our model outperforms 5.54% in negative and 0.4% in positive SNRs (even though with less SNR).

No abstract