检索结果-内蒙古大学图书馆

Parameter-efficient convolutional neural networks using wavelet transforms

AIP Conference Proceedings 2024年第1期2895卷

作者： Arnel L. Malubay Kurt Anthony C. de Los Santos Job A. Nable Ateneo de Manila University Quezon City Metro Manila Philippines

Convolutional Neural Networks (CNN's) are known to perform well on computer vision tasks such as image classification, image segmentation, and object detection. However, one major drawback of CNN's is the huge amount of computing and memory resources needed to train them. In this paper, we propose an architectural unit which we call Upsampling-Based wavelet Residual Block (UBWRB), that utilizes the 2D discrete wavelet transform coupled with upsampling operators and a residual connection to extract features from image data while having relatively fewer trainable parameters as compared to traditional convolutional layers. The discrete wavelet transform is a family of transforms that find extensive applications in signal processing and time-frequency analysis. For this paper, we use the filter-bank implementation of the discrete wavelet transform, allowing it to act in a similar fashion to a convolutional layer with fixed kernel weights. We demonstrate the performance and parameter-efficiency of CNN's with UBWRB's in the task of image classification by training them on the MNIST, Fashion-MNIST, and CIFAR-10 datasets. Our best-performing models achieve a test accuracy of 99.34% on the MNIST dataset while having less than 120,000 trainable parameters, and 92.90% and 84.27% on the Fashion-MNIST and CIFAR-10 datasets respectively, with both having less than 180,000 trainable parameters.

关键词：

来源：评论

学校读者我要写书评

暂无评论

Pavement images Denoising with Cracks Detection and Classification Using 2D Discrete wavelet Transform and Savitzky-Golay Filters

Pavement Images Denoising with Cracks Detection and Classifi...

引用

IEEE International Conference on signal and image processing applications (IEEE ICSIPA)

作者： Hamici, Zoubir Obaidat, Turki I. Al-Suleiman Al Zaytonnah Univ Jordan Elect Engn Dept Amman Jordan Al Zaytonnah Univ Jordan Civil Engn Dept Amman Jordan

ISBN: (纸本)9781728133775

image processing has gained an increased usage and impact in modern pavement networks automatic distress severity classification (DSC). DSC defines priorities and maintenance resources optimum allocation in order to achieve a cost-effective rehabilitation process. This paper presents a novel computer vision algorithm having the ability to process, isolate and evaluate the distress severity level of a pavement. A pavement color image is converted to grayscale and then processed for image denoising of the granularity and complex texture that represent and artifact in cracks edge detection. The processing is achieved by a 2D dual-tree double density wavelet transform filter banks that significantly reduces the granularity noise while preserving the pavement cracks for edge detection. The 2D wavelet FIR filters perform analysis, soft thresholding then a synthesis of the image. The second step is then an edge detection process followed by morphological filtering and labeled components size-histogram filter to isolate false edges as residuals of denoising. A final step is performed by two Savitzky-Golay filters for the detection of longitudinal and transverse alligator cracks projections. A weighted score function with multiple parameters is used for DSC.

关键词： Computer vision wavelets filter-banks Denoising Savitzky-Golay Filters Pavement Cracks

来源：评论

学校读者我要写书评

暂无评论

Single-pixel compressive imaging in shift-invariant spaces via exact wavelet frames

arXiv

引用

arXiv 2021年

作者： Vlašić, Tin Seršić, Damir Department of Electronic Systems and Information Processing University of Zagreb Faculty of Electrical Engineering and Computing Unska 3 ZagrebHR-10000 Croatia

This paper introduces a novel framework for single-pixel imaging via compressive sensing (CS) in shift-invariant (SI) spaces by exploiting the sparsity property of a wavelet representation. We reinterpret the acquisition procedure of a single-pixel camera as filtering of the observed signal with continuous-domain functions that lie in an SI subspace spanned by the integer shifts of the box function. The signal is modeled by an arbitrary SI generator whose special case is the box function, which, as we show in the paper, is conventionally used in single-pixel imaging. We propose to use separable B-spline generators which are intuitively complemented by sparsity-inducing spline wavelets. The SI models of the acquisition and the underlying signal lead to an exact discretization of an inherently continuous-domain inverse problem to a finite-dimensional problem of CS type. By solving the CS optimization problem, a parametric representation of the signal is obtained. Such a representation offers many practical advantages in image processing applications. We propose an efficient matrix-free implementation of the framework and conduct it on the standard test images and real-world measurement data. Experimental results show that the proposed framework achieves a significant improvement of the reconstruction quality relative to the conventional discretization in CS setups. MATLAB implementation of the method described in this paper has been made publicly available on https://***/retiro/compressive_imaging_in_si_spaces. © 2021, CC BY-NC-ND.

关键词： Compressed sensing

来源：评论

学校读者我要写书评

暂无评论

A quantitative assessment of speckle noise reduction in SAR images using TLFFBP neural network

引用

ARABIAN JOURNAL OF GEOSCIENCES 2020年第2期13卷 1-17页

作者： Murugesan, Kalaiyarasi Balasubramani, Perumal Rajasekaran, M. Pallikonda Kalasalingam Acad Res & Educ Dept Elect & Commun Engn Srivilliputhur Tamil Nadu India

Synthetic aperture radar (SAR) images are difficult to analyze due to the presence of speckle noise. Speckle noise must be filtered out before applying to other image processing applications. Three-layered feed forward back propagation neural network (TLFFBPNN) has been proposed to suppress the speckle noise. Gray-level co-occurrence matrix properties have been extracted, and back propagation training algorithm is used to train the neural network. The performance metrics such as peak signal to noise (PSNR), structural similarity index matrix (SSIM), edge preservation index (EPI), equivalent number of looks (ENL), and speckle suppression index (SSI) have been evaluated to find the efficiency of TLFFBPNN and compared with four recently developed de-speckling techniques. The exploratory outcomes show that the TLFFBPNN method has better de-speckling execution with great edge preservation. The comparative outcome reveals that the proposed TLFFBPNN de-speckled method outperformed in terms of PSNR of 0.98%, SSIM of 1.0%, SSI of 2.0%, EPI of 0.84%, and ENL of 0.5% when compared with the Wiener Filter Sparse Optimization in Contourlet transform domain de-speckling method.

关键词： SAR Speckle noise Contour wavelet transform Wiener filter Neural network SSI SSIM EPI ENL

来源：评论

学校读者我要写书评

暂无评论

Reversible robust data hiding based on wavelet filters modification

引用

MULTIMEDIA TOOLS AND applications 2019年第22期78卷 31847-31865页

作者： Golabi, Sasan Helfroush, Mohammad Sadegh Danyali, Habibollah Shiraz Univ Technol Dept Elect & Elect Engn Shiraz Iran

In this paper, a new robust reversible data hiding method is proposed. The method is designed based on wavelet modifications which result in a scalable data hiding scheme. The well-known biorthogonal wavelets are modified according to the watermarking bits. This is done in a way that the embedded bit can easily be interpreted based on the wavelet coefficients of the watermarked image and regardless of its resolution. Following such an algorithm would result in both reversibility and robustness. The proposed method is especially robust against wavelet resolution changing attacks and DWT based compressions. This can be of high value when dealing with low bandwidth communication situations. The practical results show high robustness against signal processing attacks and high PSNR and capacity in lossless scenarios.

关键词： Watermarking Steganography Digital wavelet transform Jpeg2000 Scalable data hiding

来源：评论

学校读者我要写书评

暂无评论

Towards the Definition of a Low-Cost Toolbox for Qualitative Inspection of Painted Historical Vaults by Means of Modified DSLR Cameras, Open Source Programs and signal processing Techniques 20th

Towards the Definition of a Low-Cost Toolbox for Qualitative...

引用

20th International Conference on Computational Science and Its applications (ICCSA)

作者： Piroddi, Luca Calcina, Sergio Vincenzo Trogu, Antonio Vignoli, Giulio UniCA DICAAR Dept Civil Engn Environm Engn & Architecture Cagliari Italy Geol Survey Denmark & Greenland GEUS Aarhus Denmark

ISBN: (纸本)9783030588205;9783030588199

Historical architecture is a primary element containing the identity values of a society. The wide diffusion of many ancient buildings gathering part of these values on painting walls over territories often characterized by poor technological or economic resources brings to consider the development of low-cost protocols to inspect valued surfaces and to give the authorities in charge of preservation and restoration adequate technical information. Here we present the preliminary results of a recent application of remote sensing micro-geophysical techniques to typical architectural targets such as vaults. A modified commercial Digital Single-Lens Reflex (DSLR) camera was used to acquire multispectral datasets on portions of a painted vault. Multispectral datasets were used raw or after the application of a pre-processing step with a Multi images Stacking (MIS) algorithm. Multispectral images were then processed with spatial wavelet decomposition, histogram enhancing, thresholds application, image fusion, false colors compositing and Principal Component Analysis (PCA) techniques. Software used have been GNU image Manipulation Program (GIMP) and Mathworks MATLAB (which can be substituted for the processing steps proposed by the built-in functions of GNU OCTAVE open-source software). Processed images were able to highlight features on vault paintings revealing details of the surface or its very shallow layers which were impossible or very difficult to distinguish in raw data. In fact, they emphasized low-visible details, differences in apparently similar finishes or pigments, cracks and probably details of surface preparation.

关键词： Multispectral analysis Digital image processing PCA Cultural heritage Historical architecture Painted walls inspection Low cost diagnostics

来源：评论

学校读者我要写书评

暂无评论

An Approximate Low-Power Lifting Scheme Using Reversible Logic

引用

IEEE ACCESS 2020年 8卷 183367-183377页

作者： Raveendran, Sithara Edavoor, Pranose J. Kumar, Nithin Y. B. Vasantha, M. H. Natl Inst Technol Veling Goa India

Haar wavelet transform is an efficacious class of wavelet transform that satisfies both symmetry and orthogonality properties which are crucial in handling boundary distortion and energy preservation in image processing applications. Such applications demand power efficient design solutions that deliver high performance. Reversible logic has emerged as a solution that incorporates logical and physical reversibility to realise low power designs. This paper presents a reversible logic based design of Haar wavelet transform and lifting scheme for Haar wavelet transform, a first in literature of reversible logic. The designs are analysed to measure the efficiency of reversible logic implementations in terms of Quantum Cost (QC), Constant Inputs (CI), Garbage Outputs (GO) and Gate Count (GC). Furthermore, this paper proposes two architectures for Reversible Approximate Full Adder (RAFA) - RAFA-1 and RAFA-2;optimised explicitly for reversible logic based implementation. The proposed architectures have 25% Error Rate (ER) and optimised QC, CI, GC and GO when compared to existing exact and approximate full adder architectures implemented using reversible logic. Functional verification of the proposed architectures are performed on FPGA using 512 x 512 image. The efficiency of the image processing application is projected in terms of Structural Similarity Index Measure (SSIM) and Peak signal to Noise Ratio (PSNR). Average SSIM and average PSNR are found to be 0.9679 and 31.81dB for RAFA-1 and 0.9696 and 32.15dB for RAFA-2 which are comparable with exact full adder based design.

关键词： Reversible logic Haar wavelet transform lifting scheme for Haar wavelet transform approximate full adders image processing

来源：评论

学校读者我要写书评

暂无评论

Atmospheric Turbulence Distortion in Video: Restoration Utilizing Sparse Analysis

Atmospheric Turbulence Distortion in Video: Restoration Util...

引用

作者： Benjamin J. Sanda Western Michigan University

学位级别：博士

The removal of atmospheric turbulence (AT) distortion in long range imaging is one of the most challenging areas of research in imaging processing with an immediate need for solutions in several applications such as in military and transportation systems. AT exacerbates distortion due to non-linear geometric blur and scintillations in long-distance images and videos, severely reducing image quality and information interpretation. AT negatively impacts both human and computer vision systems, compromising visibility essential for accurate object identiﬁcation and tracking. In this dissertation, a novel sparse analysis framework is developed to address eﬃcient AT blur and scintillation removal in video. Operating under the premise that distortion-free images should be sparse in a transform domain, the application of the dual-tree complex wavelet transform is utilized on frame bursts, allowing for a new near shift-invariant complex transform space that results in higher sparsity, higher object tracking accuracy, and better resilience against camera shake, geometric distortion, and imperfect frame registration encountered in real-world AT-distorted sequences. Using this new complex transform space the novel Frame-Burst Coeﬃcient Shimmer Thresholding (FBST) algorithm is developed. FBST considers the complex coeﬃcient shimmer across multiple frames to address threshold selection and moving object blur, issues still present in other methods which utilize techniques such as averaging and empirical threshold selection. In fact, by evaluating video sequences of moving vehicles with visible license plates, we show FBST produces up to an 85% sparse reconstruction with superior visual results compared to weighted and simple thresholding approaches while preserving object motion, reducing AT distortion, and enhancing object contrast and visibility. Moreover, compressed sensing (CS) methods to sparse AT distortion removal are also investigated through direct CS sampling of the coeﬃ

关键词： signal processing image processing computer vision image restoration object tracking atmospheric turbulence

来源：评论

学校读者我要写书评

暂无评论

An optimal wavelet-based multi-modality medical image fusion approach based on modified central force optimization and histogram matching

引用

MULTIMEDIA TOOLS AND applications 2019年第18期78卷 26373-26397页

作者： El-Hoseny, Heba M. El Kareh, Zeinab Z. Mohamed, Wael A. El Banby, Ghada M. Mahmoud, Korany R. Faragallah, Osama S. El-Rabaie, S. El-Madbouly, Essam Abd El-Samie, Fathi E. Benha Univ Dept Elect Engn Fac Engn Banha Egypt Menoufia Univ Dept Ind Elect & Control Engn Fac Elect Engn Menoufia Egypt Helwan Univ Fac Engn Dept Elect Commun & Comp Cairo Egypt Menoufia Univ Dept Comp Sci & Engn Fac Elect Engn Menoufia 32952 Egypt Menoufia Univ Dept Elect & Elect Commun Engn Fac Elect Engn Menoufia 32952 Egypt Taif Univ Dept Informat Technol Coll Comp & Informat Technol Al Hawiya 21974 Saudi Arabia

This paper introduces an optimal solution for wavelet-based medical image fusion using different wavelet families and Principal Component Ana1ysis (PCA) based on the Modified Central Force Optimization (MCFO) technique. The main motivation of this work is to increase the quality of medical fused images in order to provide correct diagnosis of diseases for the objective of optimal therapy. This can be achieved by fusing medical images of different modalities using an optimization technique based on the MCFO. The MCFO technique gives the optimum gain parameters that achieve the best fused image quality. Histogram matching is applied to improve the overall values of the Peak signal-to-Noise Ratio (PSNR), entropy, local contrast, and quality of the fused image. A comparative study is performed between the proposed algorithm, the traditional Discrete wavelet Transform (DWT), and the PCA fusion using maximum fusion rule. The proposed algorithm is evaluated subjectively and objectively with different fusion quality metrics. Simulation results demonstrate that the proposed MCFO optimized wavelet-based fusion algorithm using Haar wavelet and histogram matching achieves a superior performance with the highest image quality and clearest image details in a very short processing time.

关键词： image fusion Discrete wavelet transform (DWT) Modified central force optimization (MCFO) Histogram matchning

来源：评论

学校读者我要写书评

暂无评论

Spline and Spline wavelet Methods with applications to signal and image processing: Volume III Selected Topics

引用

2018年

作者： Amir Z Averbuch Pekka Neittaanmki Valery A Zheludev

ISBN: (纸本)9783319921228

This book provides a practical guide, complete with accompanying Matlab software, to many different types of polynomial and discrete splines and spline-based wavelets, multiwavelets and wavelet frames in signal and image processing applications. In self-contained form, it briefly outlines a broad range of polynomial and discrete splines with equidistant nodes and their signal-processing-relevant properties. In particular, interpolating, smoothing, and shift-orthogonal splines are presented.

关键词：

来源：评论

学校读者我要写书评

暂无评论

建议与咨询留下您的常用邮箱和电话号码，以便我们向您反馈解决方案和替代方法

时间限定

文献类型

馆藏选择

核心期刊

语言

文献类型

帮助

文字说明：

检索规则说明：

检索范例：

分类表

所选分类

限定检索结果

文献类型

馆藏范围

日期分布

学科分类号

主题

机构

作者

语言

请选择保存的检索档案：

请选择收藏分类：

通借通还

建议与咨询 留下您的常用邮箱和电话号码，以便我们向您反馈解决方案和替代方法

时间限定

文献类型

馆藏选择

核心期刊

语言

文献类型

帮助

文字说明：

检索规则说明：

检索范例：

分类表

所选分类

限定检索结果

文献类型

馆藏范围

日期分布

学科分类号

主题

机构

作者

语言

请选择保存的检索档案： 新增检索档案 确定 取消

请选择收藏分类： 新增自定义分类 确定 取消

通借通还

建议与咨询留下您的常用邮箱和电话号码，以便我们向您反馈解决方案和替代方法

请选择保存的检索档案：

请选择收藏分类：