Convolutional neural networks(CNN) are a subset of deep learning methods recently used widely for image segmentation. SegNet network [4] has shown interesting results for semantic segmentation, but it is designed to s...
详细信息
ISBN:
(纸本)9781728127231
Convolutional neural networks(CNN) are a subset of deep learning methods recently used widely for image segmentation. SegNet network [4] has shown interesting results for semantic segmentation, but it is designed to segment images with non-overlapped objects. However in some data translucent regions partially overlap. Having overlapped regions will cause methods not designed for overlapped objects to perform poorly or not work at all. To our knowledge no CNN has been designed yet to segment partially overlapped translucent objects. In this paper, we have designed a CNN to segment partially overlapped translucent regions. We used SegNet [4] as transfer learning for our overlapped image segmentation method. We also designed a new CNN with a simpler network for our data. Results on synthetic images give more than 95% segmentation accuracy for both methods.
Video frame interpolation is a technology that generates high frame rate videos from low frame rate videos by using the correlation between consecutive frames. Presently, convolutional neural networks (CNN) exhibit ou...
详细信息
Video frame interpolation is a technology that generates high frame rate videos from low frame rate videos by using the correlation between consecutive frames. Presently, convolutional neural networks (CNN) exhibit outstanding performance in imageprocessing and computer vision. Many variant methods of CNN have been proposed for video frame interpolation by estimating either dense motion flows or kernels for moving objects. However, most methods focus on estimating accurate motion. In this study, we exhaustively analyze the advantages of both motion estimation schemes and propose a cascaded system to maximize the advantages of both the schemes. The proposed cascaded network consists of three autoencoder networks, that process the initial frame interpolation and its refinement. The quantitative and qualitative evaluations demonstrate that the proposed cascaded structure exhibits a promising performance compared to currently existing state-of-the-art-methods.
Owing to flexibility of Unmanned Aerial Vehicles (UAVs) and high efficiency of imageprocessing technology, the combined systems become increasingly popular and important in the smart city operations. However, the app...
详细信息
ISBN:
(纸本)9781728177090
Owing to flexibility of Unmanned Aerial Vehicles (UAVs) and high efficiency of imageprocessing technology, the combined systems become increasingly popular and important in the smart city operations. However, the application scenarios of this technology, especially on the traffic system prediction and multi-vehicle information extraction, still need to be explored. Besides, vehicle's detailed attributes need to be considered when building models. The smart traffic system can be broadly divided into two parts, traffic facilities (e.g. traffic signals, signs and sensors) and participants (e.g. vehicles and pedestrians). Many related works are presented about traffic parameters measurements using UAVs. In this paper, the prediction and traffic signal system analysis through different categories of vehicles' dynamic characteristics extracted from UAVs is presented. The motivation and related work is introduced. A stochastic process framework is presented for multi-vehicle speed extraction and signal transition time distributions at a signalized intersection. Detection and tracking methods/algorithms are proposed. To verify the mathematical model, the experimental data is collected at one intersection, in the city of Singapore during peak hours. After data collection, aerial images are processed to extract information. The regression method and processed parameters help to fit the required dynamic functions for different types vehicles. The estimated distributions reflect the traffic signal transition time provided by ground truths nicely. Moreover, the future research is presented on enhancing the system prediction accuracy and robustness.
In the past few decades, medical imaging and soft computing have shown a symbolic growth in brain tumor segmentation. Research in medical imaging is becoming quite popular field, particularly in magnetic resonance ima...
详细信息
In the past few decades, medical imaging and soft computing have shown a symbolic growth in brain tumor segmentation. Research in medical imaging is becoming quite popular field, particularly in magnetic resonance images of brain tumor, because of the tremendous need of efficient and effective technique for evaluation of large amount of data. image segmentation is considered as one of the most crucial techniques for visualizing tissues in human being. In considering brain tumor image segmentation, manually with an expert, it is more likely that the errors are present in it. To automate image segmentation, we have proposed an algorithm to obtain a global thresholding value for a particular image. To find out an optimal threshold value we have used Differential Evolution algorithm embedded with OTSU method and trained neural network for future use. Proposed Methodology provides classification of the images successfully for brain tumors. Results show its efficiency over other methods.
Several methods for extracting food regions from food images use visual saliency to improve accuracy. The effectiveness of saliency detection methods for food extraction, however, has not been discussed sufficiently. ...
详细信息
Generative Adversarial Network (GAN) has been widely applied on Single image Super-Resolution (SISR) problems. However, there can be quite a variability in the results from the GAN-based methods. In some cases, the GA...
详细信息
ISBN:
(纸本)9781479981311
Generative Adversarial Network (GAN) has been widely applied on Single image Super-Resolution (SISR) problems. However, there can be quite a variability in the results from the GAN-based methods. In some cases, the GAN-based methods might cause structure distortion, which can be easily distinguished by human beings, especially for artificial structures, because the methods only focus on the perceptual quality of the whole image. On the other hand, PSNR-oriented methods can prevent structure distortion but with overly smoothed context. To overcome these problems, we propose a deep neural net refiner for SISR methods, not only improving perceptual quality but also preserving context structures. In the experiments, our model qualitatively and quantitatively performs favorably against the state-of-the-art SISR methods.
The identification and classification is important parts of the research in the field like underwater acoustic signalprocessing. Recently, deep learning technology has been utilized to achieve good performance in the...
详细信息
The identification and classification is important parts of the research in the field like underwater acoustic signalprocessing. Recently, deep learning technology has been utilized to achieve good performance in the underwater acoustic signal case. On the other side, there are still some problems should be solved. The first one is that it cannot achieve high accuracy by the dataset that is transformed into audio spectrum. The second one is that the accuracy of classification on the dataset is still low, so that, it cannot satisfy the real demand. To solve those problems, we firstly evaluated four popular spectrums (Audio Spectrum, image Histogram, Demon and LOFAR) for data preprocessing and selected the best one that is suitable for the neural networks (LeNet, ALEXNET, VGG16). Then, among these methods, we modified a neural network(LeNet) to fit the dataset that is transformed by the spectrum to improve the classification accuracy. The experimental result shows that the accuracy of our method can achieve 97.22 %, which is higher than existing methods and it met the expected target of practical application.
The removal of mixed noise is a stiff problem since the distribution of the noise cannot be predicted accurately. The most common mixed noise is the combination of Additive White Gaussian Noise (AWGN) and Impulse Nois...
详细信息
ISBN:
(纸本)9781479981311
The removal of mixed noise is a stiff problem since the distribution of the noise cannot be predicted accurately. The most common mixed noise is the combination of Additive White Gaussian Noise (AWGN) and Impulse Noise (IN). Many methods first attempt to remove IN but it might collapse the texture of the image. In this paper, we propose a new learning-based method using convolutional neural network (CNN) for removing mixed gaussian-impulse noise. Since our denoising network can remove various level of mixed noise, neither the preprocessing for removing IN nor noise-level estimation is necessary.
In recent years, the application of deep convolution neural network has achieved great success in image super-resolution (SR) fields. However, most of these deep convolution neural network methods are based on increas...
详细信息
ISBN:
(纸本)9781728136608
In recent years, the application of deep convolution neural network has achieved great success in image super-resolution (SR) fields. However, most of these deep convolution neural network methods are based on increasing network depth, width or input image size to achieve more accurate fitting, which will lead to a surplus of parameters and slow the convergence speed. Compared to target recognition tasks, the neural network for image super resolution often remove pooling layers to prevent losing some learned information. In this paper, we propose a dual-link residual network with pooling and deconvolution layers(RPDN). By adding pooling layers and corresponding deconvolution layers, the model can extract more non-adjacent pixel structure information, reduce the computation, and increase the sparsity of the model. Inspired by DenseNet, we propose a dual-link structure to achieve feature fusion of shallow and deep layers. One chain is to connect the convolution layers, and another chain is to connect the pooling layers. RPDN also uses Intergroup Connection(IC) to help gradient back propagation. Experiments on benchmark datasets shows our network can effectively reduce artifacts.
暂无评论