Extraction of discriminative features is an efficient step in any classification problem such as synthetic aperture radar (SAR) images classification. Polarimetric SAR (PolSAR) images with rich spatial features in two first dimensions and polarimetric characteristics in the third dimension are rich source of information for providing classification maps from the ground surface. By applying the spatial operators such as morphological filters by reconstruction, data dimensionality of the PolSAR is increased and needs feature reduction. In this work, median-mean and feature line embedding (MMFLE) is proposed for dimensionality reduction of the polarimetric-contextual cube in PolSAR images. MMFLE is stable with respect to outliers by utilizing the median-mean line metric. By an appropriate definition of scatter matrices, MMFLE maximizes the class separability. In addition, MMFLE is specially a superior feature reduction method when a small training set is available because it uses the feature line metric to model the data variations and generate virtual samples. With 10 training samples per class, MMFLE achieves 94.15% and 83.01% overall classification accuracy, respectively in Flevoland and SanFranciso PolSAR datasets acquired by AIRSAR.
This paper proposes face detection from blurry and noisy images robustly and efficiently using convolutional neural networks. It has also been demonstrated that the method of face detection from blurred images utilizing convolutional neural networks is superior to other methods under consideration concerning precision-recall and discontinuity and continuity scores. Face detection is the infrastructure of face recognition; also, it includes but is not limited to the following topics: traffic surveillance, stereo videos, finding a criminal from large crowds in terrorist accidents, calibrated stereo images, face alignment of images from sensors with heterologous wavelengths, driving license photos, and animations. Some difficulties of this topic include the lack of joint datasets, movements (displacements), changing expression, intensive illumination, the likelihood of overfitting in the case of employing high-dimensional data, and the presence of numerous blurs and various aspects. There are multiple face detection methods from blurry images, such as blur kernel estimation and complex Fourier coefficients of a trained neural network, exerting metrics that define arduous patches. This paper executes face detection in noisy and blurry images applying convolutional neural networks that make face detection more applicable. This is due to exploiting two techniques for blur elimination and face detection on the foundation of convolutional neural networks.
Magnetic Resonance Imaging as non-invasive imaging can produce detailed anatomical images. MRI is a time- consuming imaging technique. Several imaging techniques, like parallel imaging, have been suggested to enhance imaging speed. Compressive Sensing MRI utilizes the sparsity of MR images to reconstruct MR images with under-sampled k-space data. It has already been shown that convolutional neural networks work better than sparsity-based approaches in image quality and reconstruction speed. In this paper, a novel method based on very deep CNN for the reconstruction of MR images is proposed using Generative Adversarial Networks. Generative and discriminative networks are designed with improved ResNet architecture. Using improved architecture has led to deepening generative and discriminative networks, reducing aliasing artifacts, more accurate reconstruction of edges, and better reconstruction of tissues. Compared to DLMRI and DAGAN methods, we demonstrate the proposed method outperforms the conventional methods and deep learning-based approaches. Assessment is made on several datasets such as the brain, heart, and prostate. Reconstruction of brain data with a Cartesian mask of 30% in the proposed method has improved the SSIM criteria up to 0.99. Also, image reconstruction time is approximately 20 ms on GPU, which is suitable for real-time applications.
Cancer is a rampant phenomenon caused by uncontrollable cells that grow and spread throughout the body. Invasive Ductal Carcinoma 1 is the most common type of breast cancer, which can be fatal for females if not detected early. As a result, prompt diagnosis is critical to maximizing surveillance rates and, in the meantime, minimizing long-term mortality rates. Nowadays, modern computer vision and deep learning techniques have transformed the medical image analysis arena. Computer vision application in medical image analysis has provided us with remarkable results, enhanced accuracy, and reduced costs. The main purpose of designing a new algorithm to detect unusual patches of breast images, was to acquire both high accuracy and low computational cost, simultaneously. Therefore, a novel architecture has been designed by utilizing Xception and MobileNetV2.This new algorithm achieves 93.4% balanced accuracy and 94.8% for F1-Score, which outperforms previously published algorithms for identifying IDC histopathology images that use deep learning techniques.
In recent years, tremendous advances have been made in Artificial Intelligence (AI) algorithms in the field of image processing. Despite these advances, video compression using AI algorithms has always faced major challenges. These challenges often lie in two areas of higher processing load in comparison with traditional video compression methods, as well as lower visual quality in video content. Careful study and solution of these two challenges is the main motivation of this article that by focusing on them, we have introduced a new video compression based on AI. Since the challenge of processing load is often present in online systems, we have examined our AI video encoder in video streaming applications. One of the most popular applications of video streaming is traffic cameras and video surveillance in road environments which here we called it CCTVs. Our idea in this type of system goes back to fixed background images, where always occupied the bandwidth not efficiently, and the streaming video is related to duplicate background images. Our AI-based video encoder detects fixed background and caches it at the client-side by the background subtraction method. By separating the background image from the moving objects, it is only enough to send the moving objects to the destination, which can save a lot of network bandwidth. Our experimental results show that, in exchange for an acceptable reduction in visual quality assessment, the video compression processing load will be drastically reduced.
