Upsampling

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

Deep Space-Time Video Upsampling Networks

Abstract: Video super-resolution (VSR) and frame interpolation (FI) are traditional computer vision problems, and the performance have been improving by incorporating deep learning recently. In this paper, we investigate the problem of jointly upsampling videos both in space and time, which is becoming more important with advances in display systems. One solution for this is to run VSR and FI, one by one, independently. This is highly inefficient as heavy deep neural networks (DNN) are involved in each solution. To this end, we propose an end-to-end DNN framework for the space-time video upsampling by efficiently merging VSR and FI into a joint framework. In our framework, a novel weighting scheme is proposed to fuse all input frames effectively without explicit motion compensation for efficient processing of videos. The results show better results both quantitatively and qualitatively, while reducing the computation time (\(\times \)7 faster) and the number of parameters (30%) compared to baselines. Our source code is available at https://github.com/JaeYeonKang/STVUN-Pytorch.

Upsampling speckle patterns for coherent X-ray diffraction imaging

Abstract: Coherent X-ray diffraction imaging is a lensless imaging technique where an iterative phase-retrieval algorithm is applied to the speckle pattern, the far-field diffraction pattern produced by an isolated object. To ensure convergence to a unique solution, the diffraction pattern must be oversampled by a factor of two or more. Since the resolution in real space depends on the maximum wave vector where the intensity is detected, i.e. on the detector field of view, there is a practical limitation on oversampling in reciprocal space and resolution in real space that is ultimately determined by the number of pixels. This work shows that it is possible to reduce the effective pixel size and maintain the detector field of view by applying a linear combination method to shifted diffraction patterns. The feasibility of the method is demonstrated by reconstructing the images of test objects from diffraction patterns oversampled in each dimension by factors of 1.3 and 1.8 only. The described approach can be applied to any diffraction or imaging technique where the resolution is compromised by a large pixel size.

AutoSegNet: An Automated Neural Network for Image Segmentation

Abstract: Neural Architecture Search (NAS) has drawn significant attention as a tool for automatically constructing deep neural networks. The generated neural networks are mainly applied for image classification, and natural language processing. However, there are increasing demands for image segmentation in various areas, such as medical image processing, satellite image object location, and autopilot technology. We propose a NAS method called Automated Segmentation Network (AutoSegNet), targeting industrial and medical image segmentation. The search architectures are constructed by stacking the downsampling layer, the bridge layer, and the upsampling layer, which are explored by a recurrent neural network. Compared with other related methods for image segmentation, the proposed method has a small search space but can explore most of the-state-of-the-art supervised image segmentation models. We perform verification on two datasets, and the results show that AutoSegNet achieves superior segmentation results with clear and continuous segmented edges, as well as better image details.

Design and Simulation of a Hybrid Filter Bank for Processing Wideband Signals with Low-Speed ADCs

Abstract: Currently, there is a tendency to increase the speed of information transmission by increasing used frequency band. However, when switching to broadband signals and operating under conditions of a high level of noise, a wide dynamic range is required when digitizing the received signal. To increase the dynamic range, it is proposed to use a hybrid filter bank, which allows processing the broadband signal with low-speed precision analog-to-digital converters. In this paper, was calculated and simulated a hybrid analysis-synthesis filter bank for the 26.25 MHz-86.25 MHz frequency range. Using matched filters, we estimated the distortion of a linearly frequency-modulated signal during the passage of filter banks without intermediate processing, after upsampling and downsampling intermediate signals from the outputs of analog filters. To reduce phase shifts and time delays arising during signal processing by synthesis filters, compensation delays and phase shifts were introduced, adjusted to distortions carried by the processing system. It was shown that phase correction allows reducing signal distortions. It was also shown that in the case of receiving a mixture of signal and noise, the distortion introduced by the processing system does not significantly affect the signal-to-noise ratio of the output signal.

On the design and multiplier-less realization of digital IF for software radio receivers with prescribed output accuracy

Abstract: This paper studies the design and multiplier-less realization of the digital IF in software radio receivers The new architecture consists of a compensator for compensating the passband droop of the conventional cascaded integrator and comb (CIC) filter The passband droop is improved by a factor of four and it can be implemented with four additions using the sum-of-powers-of-two (SOPOT) coefficients The decimation factor of the multistage decimator is also reduced so that its output can be fed directly to the Farrow structure for sample rate conversion (SRC), eliminating the need for another L-band filter for upsampling By so doing, the programmable FIR filter can be replaced by a half-band filter placed immediately after the Farrow structure As the coefficients of this half-band filter, the multistage decimators and the subfilters in the Farrow structure are constants, they can be implemented without multiplication using SOPOT coefficients As a result, apart from the limited number of multipliers required in the Farrow structure, the entire digital IF can be implemented without any multiplications A random search algorithm is employed to minimize the hardware complexities of the proposed IF subject to a given specification in the frequency domain and prescribed output accuracy, taking into account signal overflow and round-off noise Design results are given to demonstrate the effectiveness of the proposed method