The current superresolution (SR) methods based on deep learning have shown remarkable comparative advantages but remain unsatisfactory in recovering the high-frequency edge details of images noise-contaminated imaging conditions, e.g., remote sensing satellite imaging. In this paper, we propose a generative adversarial network (GAN)-based edge-enhancement (EEGAN) for robust image SR reconstruction along with strategy that is insensitive to noise. particular, EEGAN consists two main...

Video super-resolution (SR) is focused on reconstructing high-resolution frames from consecutive low-resolution (LR) frames. Most previous video SR methods based convolutional neural networks (CNN) use a direct connection and single-memory module within the network, thus, they fail to make full of spatio-temporal complementary information LR observed To fully exploit correlations between adjacent reveal more realistic details, this paper proposes multi-memory CNN (MMCNN) for SR, cascading an...

Purpose This paper presents a novel hands‐free control system for intelligent wheelchairs (IWs) based on visual recognition of head gestures. Design/methodology/approach A robust gesture‐based interface (HGI), is designed gesture the RoboChair user. The recognised gestures are used to generate motion commands low‐level DSP controller so that it can according user's intention. Adaboost face detection algorithm and Camshift object tracking combined in our achieve accurate detection, real time....

The goal of learning-based image super resolution (SR) is to generate a plausible and visually pleasing high-resolution (HR) from given low-resolution (LR) input. SR problem severely underconstrained, it has rely on examples or some strong priors reconstruct the missing HR details. This paper addresses learning mapping functions (i.e., projection matrices) between LR images based dictionary examples. Encouraged by recent developments in prior modeling, where state-of-the-art algorithms are...

Recently, the application of satellite remote sensing images is becoming increasingly popular, but observed from sensors are frequently in low-resolution (LR). Thus, they cannot fully meet requirements object identification and analysis. To utilize multi-scale characteristics objects images, this paper presents a residual neural network (MRNN). MRNN adopts nature to reconstruct high-frequency information accurately for super-resolution (SR) imagery. Different sizes patches LR initially...

How to effectively fuse temporal information from consecutive frames remains be a non-trivial problem in video super-resolution (SR), since most existing fusion strategies (direct fusion, slow or 3D convolution) either fail make full use of cost too much calculation. To this end, we propose novel progressive network for SR, which are processed way separation and the thorough utilization spatio-temporal information. We particularly incorporate multi-scale structure hybrid convolutions into...

Rain streaks in the air show diverse characteristics with different shapes, directions, densities, even complex overlapped phenomenon, causing great challenges for deraining task. Recently, deep learning based image methods have been extensively investigated due to their excellent performance. However, most of existing algorithms still limitations removing rain while preserving rich textural details under complicated conditions. To this end, we propose decompose into multiple layers and...

Most recent video super-resolution (SR) methods either adopt an iterative manner to deal with low-resolution (LR) frames from a temporally sliding window, or leverage the previously estimated SR output help reconstruct current frame recurrently. A few studies try combine these two structures form hybrid framework but have failed give full play it. In this paper, we propose omniscient not only utilize preceding output, also outputs present and future. The is more generic because iterative,...

Recently, convolutional neural networks (CNNs) have been widely employed to promote the face hallucination due ability predict high-frequency details from a large number of samples. However, most them fail take into account overall facial profile and fine texture simultaneously, resulting in reduced naturalness fidelity reconstructed face, further impairing performance downstream tasks (e.g., detection, recognition). To tackle this issue, we propose novel external-internal split attention...

Low-light image enhancement aims to improve an image's visibility while keeping its visual naturalness. Different from existing methods, which tend accomplish the relighting task directly, we investigate intrinsic degradation and relight low-light refining details color in two steps. Inspired by formulation (diffuse illumination plus environment color), first estimate inputs simulate distortion of color, then refine content recover loss diffuse color. To this end, propose a novel...

In recent years, remote sensing images have attracted a lot of attention because their special value. However, acquired by satellite sensors are usually low-resolution (LR), so much more difficult to infer high-frequency details from compared with ordinary digital images, which means they cannot meet the needs certain downstream tasks. this letter, we propose multiscale enhancement network (MEN), uses features enhance network's reconstruction capability. Specifically, extracts coarse LR...

Fog removal from an image is active research topic in computer vision. However, current literature weak the following two areas which many ways are hindering progress for developing defogging algorithms. First, there no true real-world and naturally occurring foggy datasets suitable models. Second, mathematically simple easy to use quality assessment (IQA) methods evaluating visual of defogged images. We address these aspects this paper. first introduce a new dataset called multiple (MRFID)....

As a fundamental and critical task in feature-based remote sensing image registration, feature matching refers to establishing reliable point correspondences from two images of the same scene. In this article, we propose simple yet efficient method termed linear adaptive filtering (LAF) for both rigid nonrigid apply it registration task. Our algorithm starts with putative based on local descriptors then focuses removing outliers using geometrical consistency priori together denoising theory....

Abstract A generic intelligent tomato classification system based on DenseNet-201 with transfer learning was proposed and the augmented training sets obtained by data augmentation methods were employed to train model. The trained model achieved high accuracy images of different quality, even those containing levels noise. Also, could accurately efficiently identify classify a single image only 29 ms, indicating that has great potential value in real-world applications. feature visualization...

Most existing deep learning-based pan-sharpening methods own several widely recognized issues, such as spectral distortion and insufficient spatial texture enhancement. To address these challenges in pan-sharpening, we propose a novel dual-path fusion network (DPFN). The proposed DPFN includes two major components: 1) the global subnetwork (GSN) 2) local (LSN). In particular, GSN aims to search similar image blocks panchromatic (PAN) space multispectral (MS) exploits HR textural information...

Recently, face hallucination methods either feed whole image into convolutional neural networks (CNNs) or utilize extra facial priors (e.g., parsing maps and landmarks) to focus on global structure constrain texture generation. However, the limited receptive fields of CNNs inaccurate will reduce naturalness fidelity restored face. In this paper, we propose a FaceFormer that aggregates representation Transformers local maintain consistency while restoring details. The reason for design is...

Remote sensing images exhibit rich texture features and strong autocorrelation. Although the super-resolution (SR) method of remote based on convolutional neural networks (CNN) can capture local information, limited perceptual field prevents it from establishing long-distance dependence global leading to low accuracy image reconstruction. Furthermore, is difficult for existing SR methods be deployed in mobile devices due their large network parameters high computational demand. In this...