Upsampling

About: Upsampling is a research topic. Over the lifetime, 2426 publications have been published within this topic receiving 57613 citations.

Single photograph super resolution enhancement method based on depth residual network

Abstract: The invention provides a single photograph super resolution enhancement method based on a depth residual network. The main content comprises convolution modules, an independent structure layer, a transformation model and a loss function. The process comprises the steps of using a series of convolution modules and pixel reconstruction modules to build the depth residual network, using the independent convolution layer and data superposition layer to carry out refinement and superposition of image residues, allowing intermediate data to pass the pixel reconstruction module, and improving the clarity of an input image by an upsampling operation inside. According to the method, the super resolution enhancement of images of different sizes can be processed, single-transform and multi-transformframeworks are provided to solve image generation, and at the same time, the definition and resolution of the image are improved under the control of a distortion degree of image vision.

Singular Lorenz Measures Method for seizure detection using KNN-Scatter Search optimization algorithm

Abstract: Offline algorithm to detect the intractable epileptic seizure of children has vital role for surgical intervention. In this paper, after preprocessing and windowing procedure by Discrete Wavelet Transform (DWT), EEG signal is decomposed to five brain rhythms. These rhythms are formed to 2D pattern by upsampling idea. We have proposed a novel scenario for feature extraction that is called Singular Lorenz Measures Method (SLMM). In our method, by Chan's Singular Value Decomposition (Chan's SVD) in two phases including of QR factorization and Golub-Kahan-Reinsch algorithm, the singular values as energies of the signal on orthogonal space for pattern of rhythms in all windows are obtained. The Lorenz curve as a depiction of Cumulative Distribution Function (CDF) of singular values set is computed. With regard to the relative inequality measures, the Lorenz inconsistent and consistent features are extracted. Moreover, the hybrid approach of K-Nearest Neighbor (KNN) and Scatter Search (SS) is applied as optimization algorithm. The Multi-Layer Perceptron (MLP) neural network is also optimized on the hidden layer and learning algorithm. The optimal selected attributes using the optimized MLP classifier are employed to recognize the seizure attack. Ultimately, the seizure and non-seizure signals are classified in offline mode with accuracy rate of 90.0% and variance of MSE 1.47×10−4.

Design and architecture of multiplier-free FIR filters using periodically time-varying ternary coefficients

Abstract: Multiplier-free realizations for FIR filters are proposed. The realizations use a periodically time-varying (PTV) system, flanked by simple units for upsampling and downsampling to achieve time-invariant multiplier-free FIR filter operation. The PTV system uses only ternary (0, +or-1) coefficients, and the units before and after the PTV system use only power-of-two scalers. Therefore, the realizations can be implemented with only add/subtract operations. Some architectures for the proposed structures are also presented. >

NeuralSampler: Euclidean Point Cloud Auto-Encoder and Sampler.

Abstract: Most algorithms that rely on deep learning-based approaches to generate 3D point sets can only produce clouds containing fixed number of points. Furthermore, they typically require large networks parameterized by many weights, which makes them hard to train. In this paper, we propose an auto-encoder architecture that can both encode and decode clouds of arbitrary size and demonstrate its effectiveness at upsampling sparse point clouds. Interestingly, we can do so using less than half as many parameters as state-of-the-art architectures while still delivering better performance. We will make our code base fully available.

Infrared Small Target Detection Network With Generate Label and Feature Mapping

Abstract: In the complex background, the contrast and signal-to-noise (SNR) ratio of the infrared (IR) small target are low. Therefore, the traditional IR small target detection algorithms are difficult to achieve good detection performance when the characteristics of small targets are sparse. To solve this problem, an IR small target detection network with generate label and feature mapping (GLFM)-net is proposed in this letter. First, in the GLFM-net model, a scale adaptive feature extraction network is proposed for the IR small target sparse features extraction, and then, the multilayer joint upsampling feature mapping network is proposed for small target feature mapping and background suppression. Based on this model, the feature mapping results of IR dim and small targets with the greatly suppressed background are obtained. Second, in model training, we designed a 2-D Gaussian label generation strategy for the problem of sample imbalance, which can achieve excellent detection performance by using small training samples. The experimental results show that the network can detect IR small targets with different sizes and low SNRs in various complex backgrounds and has good effectiveness and robustness compared with the existing algorithms.