稀疏补偿下过度完整补丁的高效两步面部修复法

51 浏览量更新于2024-08-27 收藏 717KB PDF 举报

本篇研究论文探讨了一种基于稀疏补偿的过分完整补丁的有效两步幻觉方法，发表在《IET Image Processing》上，于2011年10月4日首次接收修订，并在2013年2月14日被接受。该文章的DOI为10.1049/iet-ipr.2012.0554，其ISSN为1751-9659。作者是Mohamed Naleer Haju Mohamed和Yao Lu，他们分别来自北京理工大学计算机科学学院的北京智能信息技术实验室和斯里兰卡东南大学的应用科学学院。研究的核心内容聚焦在利用低分辨率和高分辨率人脸图像进行图像处理，特别是面部幻觉技术。首先，通过插值技术得到训练图像的最佳系数，构建出全局的面部图像。这种方法强调了稀疏表示在图像重构中的关键作用，因为通过找到图像中各个部分的最简表示，可以有效地保留图像的本质特征。接着，作者采用过完备的补丁字典，对局部的高分辨率剩余图像（即所谓的局部面部图像）进行检测和识别。过完备字典意味着它包含远超过实际数据所需的基元素，这有助于捕获更丰富的图像特征并提高恢复精度。通过这种字典，研究人员能够找到最匹配的稀疏表示，进一步细化局部面部细节。最后，通过对全局图像和局部残留图像的结合，实现了一种高效的面部幻觉过程，生成出具有高质量、细节清晰的幻象面部图像。这种方法的优势在于它能够在保持整体一致性的同时，增强局部的精细特征，从而在视觉上达到令人信服的效果。这篇论文提出了一种创新的策略，将稀疏补偿和过完备补丁技术融合到面部幻觉中，旨在提升图像处理的精确性和逼真度，对于图像重建和超分辨率技术的研究具有重要的理论价值和实践意义。

Published in IET Image Processing

Received on 4th October 2011

Revised on 12th January 2013

Accepted on 14th February 2013

doi: 10.1049/iet-ipr.2012.0554

ISSN 1751-9659

Effective two-step method for face hallucination based

on sparse compensation on over-complete patches

Mohamed Naleer Haju Mohamed

1,2

, Yao Lu

, Feng Lv

Beijing Laboratory of Intelligent Information Technology, School of Computer Science, Beijing Institute of Technology,

Beijing 100081, People’s Republic of China

Faculty of Applied Sciences, South Eastern University of Sri Lanka, Amparai 3220, Sri Lanka

E-mail: hmmnaleer@gmail.com

Abstract: Sparse representation has been successfully applied to image d using low- and high-resolution training face images

based on sparse representation. In this study, the sparse residual compensation is adopted to face hallucination. Firstly, a

global face image is constructed by optimal coefﬁcients of the interpolated training images. Secondly, the high-resolution

residual image (local face image) is found by using an over-complete patch dictionary and the sparse representation. Finally,

a hallucinated face image is obtained by combining these two steps. In addition, the more details of the face image in high

frequency parts are recovered using a residual compensation strategy. In the authors’ experimental work, it is observed that

balance sparsity parameter (λ) has affected the residual compensation. Further, the proposed algorithm can acquire a high-

resolution image even though the number of training image pairs is comparatively smaller. The experiments show that the

authors’ method is more effective than the other existing two-step face hallucination methods.

1 Introduction

In the past few years, many researchers focused on sparse

representation and compressive sensing. Survey on the

theory and algorithms of compressed sensing and sparse

representation has been seen in [1–4]; Donoho and Tanner

[1] provide an elegant geometric interpretation of classical

compressed sensing theory and results, revealing a strong

connection to high-dimensional combinatorial geometry and

providing a precise characterisation of under-sampling

bounds. Cande’s and Plan [2] introduce a more recent trend

of extending the study from recovery of sparse signals to

the completion of low-rank matrices, in which the

sparsity-prompting l

-norm is replaced with the low-rank

promoting nuclear norm of a matrix. This work proves the

stability of matrix completion under noise.

Tropp and Wright [3] give a comprehensive review of many

algorithms that had been developed in the past few years for

sparse signal recovery. Cevher et al. [4] demonstrate how

the basic sparse signal model can be extended or generalised

to broader classes of low-dimensional models with similar

efﬁcient and accurate signal reconstruction. These models

are applicable to wider range of applications. Some of the

conventional applications of compressive sensing in signal

processing, including images, are stated in [5, 6] in detail.

The surveys by Elad and Aharon [5] provide both

conceptual and empirical justiﬁcations why sparse and

redundant representations are crucial for image processing.

Sparsity-promoting techniques are very effective for many

classical tasks such as image denoising, inpainting,

super-resolution etc. Fadili et al.[6] apply sparse

representation to more speciﬁc image processing problem of

decomposing image into multiple unknown components.

Sparsity promoting and compressive sensing techniques

have tremendous effect on a broader range of engineering

ﬁelds, including but not limited to pattern recognition,

machine learning, communications, sensor networks and

image processing [7–10]; Wright et al.[7] survey some of

the early successful applications of sparse representations in

computer vision, especially in face recognition. It shows

that special settings of the practical problems lead to new

mathematical models and results that enrich the basic theory

of sparse representation. Rubinstein et al.[8] address the

important problem how to automatically learn a dictionary

from raw samples that are adaptive to the signals or data of

interest. Also, a comprehensive survey of some of the most

popular and effective algorithms for learning dictionary has

been stated by Bajwa et al.[9] shows how compressed

sensing can be used to estimate the parameters of the

communication channels. Yang et al.[10] demonstrate how

sparse representation can be used for event detection and

classiﬁcation with sensor netw orks.

Recently, the super-resolution is a very active research area.

It aims to overcome some of the inherent resolution

boundaries of low-cost imaging sensor. Such resolution

enhancing skill may also be crucial in judgement or

investigation from low-quality images and the related

practical techniques can also be extremely complicated.

Inspection systems is yet another application that is being

paid more attention, it can provide very signiﬁcant

information about the tracked objects such as criminals and

license plates.

www.ietdl.org

624 IET Image Process., 2013, Vol. 7, Iss. 6, pp. 624–632

The Institution of Engineering and Technology 2013 doi: 10.1049/iet-ipr.2012.0554

下载后可阅读完整内容，剩余8页未读，立即下载

weixin_38518006

粉丝: 3
资源: 996

稀疏补偿下过度完整补丁的高效两步面部修复法

基于稀疏分解的图像去噪

基于稀疏局部像素结构的人脸幻觉

基于逐步稀疏重构的面部幻觉

带有基于稀疏补丁的表示的改进标签融合方法用于MRI脑图像分割

论文研究-基于稀疏数据规约的CMAC大气腐蚀数据补偿方法.pdf

基于稀疏表示的超分辨率方法

基于稀疏表示的3D模型压缩方法

基于稀疏表达的图像去噪方法研究

通过径向基函数-偏最小二乘回归和局部性稀疏表示的监控人脸图像两步超分辨率方法

基于稀疏表示的电力设备图像识别方法

最新资源