A5109910272- Image and Signal Denoising Methods
- Advanced Data Compression Techniques
- Digital Filter Design and Implementation
- Image Processing Techniques and Applications
- Advanced Image Processing Techniques
- Sparse and Compressive Sensing Techniques
- Advanced Vision and Imaging
- Advanced Image Fusion Techniques
- Advanced Image and Video Retrieval Techniques
- Algorithms and Data Compression
- Advanced Wireless Communication Techniques
- Optical Systems and Laser Technology
- Image Retrieval and Classification Techniques
- Error Correcting Code Techniques
- Medical Image Segmentation Techniques
- Face recognition and analysis
- Face and Expression Recognition
- Video Coding and Compression Technologies
- Medical Imaging Techniques and Applications
- Image and Video Stabilization
- Numerical Methods and Algorithms
- Infrared Target Detection Methodologies
- Cryptography and Residue Arithmetic
- Advanced Measurement and Detection Methods
- Advanced Algorithms and Applications
Minzu University of China
2023-2024
ETH Zurich
2022
University of Nottingham Ningbo China
2022
South Central Minzu University
2012-2021
ORCID
2019
Huazhong University of Science and Technology
2003-2008
This paper reviews the NTIRE 2022 challenge on efficient single image super-resolution with focus proposed solutions and results. The task of was to super-resolve an input a magnification factor ×4 based pairs low corresponding high resolution images. aim design network for that achieved improvement efficiency measured according several metrics including runtime, parameters, FLOPs, activations, memory consumption while at least maintaining PSNR 29.00dB DIV2K validation set. IMDN is set as...
Novel architectures for 1-D and 2-D discrete wavelet transform (DWT) by using lifting schemes are presented in this paper. An embedded decimation technique is exploited to optimize the architecture DWT, which designed receive an input generate output with low- high-frequency components of original data being available alternately. Based on DWT architecture, efficient line-based further proposed employing parallel pipeline techniques, mainly composed two horizontal filter modules one vertical...
A fast and efficient hardware implementation for computing the Singular value decomposition (SVD) Eigenvalue (EVD) is presented. Considering that SVD EVD are complex expensive operations, to achieve high performance with low complexity, our approach takes full advantage of combination parallel sequential computation, which can increase efficiently utilization. Besides, regarding EVD, we propose a solution simplified Coordinate rotation digital computer (CORDIC)-like algorithm obtain higher...
Non-local similarity-based group sparse representation (GSR) has shown great potential in image restoration. Considering the universal existing non-stationarity of natural images and statistic characteristic differences different components domain patch group, this paper proposes a new compressive sensing reconstruction (ICSR) algorithm based on z-score standardized (ZSGSR). Specifically, is first partitioned into overlapping patches, similar groups are further generated to be decomposed by...
A flipping structure for the discrete wavelet transform (DWT) was proposed by Huang et al., which aimed at shortening critical path of lifting-based one-dimensional (1-D) architecture and reducing number pipeline register used in 1-D as well size temporal buffer (TB) required line-based two-dimensional (2-D) architecture. In this paper, we explore intrinsic relationship between TB 2-D DWT (LBA2DDWT) registers a module, present modified view on issue that has been al. An improved method...
A novel approach of joint gravity and gradient inversion is presented for passive subsurface object detection in this letter. Gravity anomalies induced by an can be measured inversed to estimate the mass, orientation, distance object. The new equations are explored, weighted least squares estimation constructed. Therefore, two kinds data complementary each other via self-adaptive weights. Simulation results show that proposed more efficient robust than previous tensor inversion.
Few-shot segmentation (FSS) is a challenging task because the same class of targets in support and query image may have different scale, texture background information. Prototype learning (PL) current mainstream FSS method, which characterizes interaction between prototype vector feature. However, commonly based on global average pooling only contains first-order feature information, vulnerable to varying appearance similar target diversity background. Moreover, auxiliary information not...
Aiming to the issue of face recognition with partial contiguous occlusion, a new method was proposed by removing outlier area in this paper. A mean image is firstly obtained from train images, which subtracted test form an error image. Then used obtain occlusion segmentation technique, and images are tailored corresponding area. Finally, performed linear regression classifier or sparse coding classifier. Compared similar works, has considerably performance improvement relatively simple...
The different image block has sparsity or compressibility in transform domain; general, the blocks smooth region have stronger while those texture edge weaker sparsity. Based on this observation, a novel DCT based sampling scheme with coefficients random permutations for compressive sensing been proposed paper. These make of all sampled more evenly, which results requiring approximate equal ratio measurement well reconstruction each block. Experimental demonstrate that our can efficiently...
Block-based random image sampling using compressed sensing needs a lot of measurements which increases the complexity compression process. This paper proposes novel nonuniform method to measure discrete cosine transform coefficients blocks. The in zigzag order are divided into important and unimportant frequency component differentiated Taking this advantage, proposed algorithm makes it possible achieve remarkable performance gain at same number measurements. Experiments have demonstrated...
The embedded block coding with optimized truncation (EBCOT) is the state-of-the-art technique for image compression, which heart of latest still compression standard JPEG2000. EBCOT can be partitioned into three steps: bit plane (BPC), context adaptive binary arithmetic (CABAC) and post-compression rate-distortion (PCRD) optimization. in JPEG2000 a fractional bit-plane method, transformed coefficients code are coded by 3 passes each bit-plane, context-decision (CX-D) pair symbol generated...
Recently, facial priors based face super-resolution (SR) methods have obtained significant performance gains in dealing with extremely degraded images, and also been proved useful facilitating the inference of images. Based on this, how to efficiently fuse into deep features improve SR has attracted a major attention. In this paper, we propose learning-based approach Incremental boosting Parsing information (dubbed IPFSR) for high magnification factor. The proposed IPFSR method consists...
目的 基于深度学习的图像超分辨率重构研究取得了重大进展,如何在更好提升重构性能的同时,有效降低重构模型的复杂度,以满足低成本及实时应用的需要,是该领域研究关注的重要问题。为此,提出了一种基于通道注意力(channel attention,CA)嵌入的Transformer图像超分辨率深度重构方法(image super-resolution with channelattention-embedded Transformer,CAET)。方法 提出将通道注意力自适应地嵌入Transformer变换特征及卷积运算特征,不仅可充分利用卷积运算与Transformer变换在图像特征提取的各自优势,而且将对应特征进行自适应增强与融合,有效改进网络的学习能力及超分辨率性能。结果 基于5个开源测试数据集,与6种代表性方法进行了实验比较,结果显示本文方法在不同放大倍数情形下均有最佳表现。具体在4倍放大因子时,比较先进的SwinIR (image restoration using swin Transformer)方法,峰值信噪比指标在Urban100数据集上得到了0.09...
The speed of MR acquisition can be increased by decreasing the number sequential phase encodes means arrays multiple receiver coils. However, coil data are frequently corrupted due to involuntary movement during acquisition, which in turn have a significant influence on reconstructed composite image. In this article, new method called Weighted-Sum-of-Squares (WSOS) is proposed remove artifacts images. firstly adopts homomorphic filtering and B-spline smoothing obtain group uniform brightness...
This paper proposes a novel fast architecture for 2D discrete wavelet transform by using lifting scheme, Parallel and embedded decimation techniques are employed to optimize the architecture, which is mainly composed of two horizontal filter modules one vertical module, working in parallel pipeline fashion with 100% hardware utilization. The designed generate outputs clock cycle, every subbands coefficients alternately. total time computing J levels decomposition an N/spl times/N image...
MQ arithmetic coder has been adopted to achieve entropy coding in the latest image compression standard JPEG2000. which is a bit-level operation with intensive branch and feedback thus becomes serious bottleneck of high speed In this paper, an efficient fast algorithm for was proposed, renormalization process BYTEOUT performed batch fashion instead gradual iteration as introduced Experimental results have proved validity method decreasing computation complexity.