Upsampling

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

Wavelet Codes for Algorithm-Based Fault Tolerance Applications

Abstract: Algorithm-based fault tolerance (ABFT) methods, which use real number parity values computed in two separate comparable ways to detect computer-induced errors in numerical processing operations, can employ wavelet codes for establishing the necessary redundancy. Wavelet codes, one form of real number convolutional codes, determine the required parity values in a continuous fashion and can be intertwined naturally with normal data processing. Such codes are the transform coefficients associated with an analysis uniform filter bank which employs downsampling, while parity-checking operations are performed by a syndrome synthesis filter bank that includes upsampling. The data processing operations are merged effectively with the parity generating function to provide one set of parity values. Good wavelet codes can be designed starting from standard convolutional codes over finite fields by relating the field elements with the integers in the real number space. ABFT techniques are most efficient when employing a systematic form and methods for developing systematic codes are detailed. Bounds on the ABFT overhead computations are given and ABFT protection methods for processing that contains feedback are outlined. Analyzing syndromes' variances guide the selection of thresholds for syndrome comparisons. Simulations demonstrate the detection and miss probabilities for some high-rate wavelet codes.

CloudU-Netv2: A Cloud Segmentation Method for Ground-Based Cloud Images Based on Deep Learning

Abstract: Accurately acquiring cloud information through cloud images segmentation is of great importance for weather forecasting, environmental monitoring, sites selection of observatory and analysis of climate evolution. In this paper, a cloud segmentation method based on deep learning, called CloudU-Netv2, is proposed to segment daytime and nighttime ground-based cloud images. The CloudU-Netv2 includes encoder, dual attention modules and decoder. The contributions in this paper are four folds as follows. Firstly, it replaces the ‘upsampling’ in CloudU-Net with ‘bilinear upsampling’. Secondly, position and channel attention modules are added to the structure to improve the discrimination ability of features’ representation. Thirdly, it chooses rectified Adam as the optimizer in the CloudU-Netv2 structure. Finally, we conduct ablation experiments on the key components of CloudU-Netv2 and compare with the existing four advanced methods using six evaluation metrics. Results show that the key components of the model play the pivotal role in improving the segmentation performance, and the proposed CloudU-Netv2 has the best segmentation performance for daytime and nighttime ground-based cloud images compared with four other methods.

Deep convolutional neural network with self-transfer learning

Abstract: Systems and techniques for facilitating a deep convolutional neural network with self-transfer learning are presented. In one example, a system includes a machine learning component, a medical imaging diagnosis component and a visualization component. The machine learning component generates learned medical imaging output regarding an anatomical region based on a convolutional neural network that receives medical imaging data. The machine learning component also performs a plurality of sequential downsampling and upsampling of the medical imaging data associated with convolutional layers of the convolutional neural network. The medical imaging diagnosis component determines a classification and an associated localization for a portion of the anatomical region based on the learned medical imaging output associated with the convolutional neural network. The visualization component generates a multi-dimensional visualization associated with the classification and the localization for the portion of the anatomical region.

Analysis and comparison of various image downsampling and upsampling methods

Abstract: Summary form only given. The goal is to gain a better understanding of the behavior of the image down/upsampling combinations, and find better down/upsampling methods. We examined existing down/upsampling methods and proposed new ones. We formulated a frequency response approach for understanding and evaluating down/upsampling combinations. The approach was validated experimentally by running the methods on various images and computing the signal to noise ratio (SNR) between the original and the down-then-upsampled images. The frequency response based evaluation correlates well with the experimental evaluation. Down/upsampling combinations were studied in a unified framework. Signals are pre-filtered then decimated by two, resulting in downsampling by two. Afterwards, signals are zero-upsampled by 2, i.e., inserting 0s between successive samples, and then post-filtering. Our analysis showed that for optimal performance, the pre-filter and the post-filter should both be low-pass filters with cutoff at /spl pi//2. We considered five classes of filters. The first corresponds to the simplest down/upsampling combination, decimation/duplication, where decimation is simply the skipping of every other row and every other column, and duplication (for upsampling) involves duplicating every row and every column. The second class corresponds to bilinear interpolation, for both upsampling and downsampling. The third class comprises the biorthogonal and orthogonal wavelets. The fourth class we termed binomial filters. The fifth class consists of least-square FIR filters.

Super-resolution of facial images in forensics scenarios

Abstract: Forensics facial images are usually provided by surveillance cameras and are therefore of poor quality and resolution Simple upsampling algorithms can not produce artifact-free higher resolution images from such low-resolution (LR) images To deal with that, reconstruction-based super-resolution (SR) algorithms might be used But, the problem with these algorithms is that they mostly require motion estimation between LR and low-quality images which is not always practical To deal with this, we first simply interpolate the LR input images and then perform motion estimation The estimated motion parameters are then used in a non-local mean-based SR algorithm to produce a higher quality image This image is further fused with the interpolated version of the reference image via an alpha-blending approach The experimental results on benchmark datasets and locally collected videos from surveillance cameras, show the outperformance of the proposed system over similar ones