High-volume data hiding in images Introducing perceptual criteria into quantization based e

HIGH-VOLUME DATA HIDING IN IMAGES:INTRODUCING PERCEPTUAL CRITERIA INTO QUANTIZATION BASED EMBEDDING

K.Solanki,N.Jacobsen,S.Chandrasekaran,U.Madhow,and B.S.Manjunath

Dept.of Electrical and Computer Engineering

University of California at Santa Barbara

Santa Barbara,CA93106

ABSTRACT

Information-theoretic analyses for data hiding prescribe em-bedding the hidden data in the choice of quantizer for the host data.In this paper,we consider a suboptimal implementation of this prescription,with a view to hiding high volumes of data in images with low perceptual degradation.Our two main?ndings are as follows:

(a)In order to limit perceptual distortion while hiding large amounts of data,the hiding scheme must use perceptual criteria in addition to information-theoretic guidelines.

(b)By focusing on“benign”JPEG compression attacks,we are able to attain very high volumes of embedded data,comparable to information-theoretic capacity estimates for the more malicious Additive White Gaussian Noise(AWGN)attack channel,using relatively simple embedding techniques.

1.INTRODUCTION

Data hiding is the process by which a message signal,or signa-ture,is covertly embedded within a host data set to form a com-posite signal.A well-accepted application of data hiding is for watermarking,which requires embedding a relatively short string of data in the host data(e.g.,for identifying the source/copyright for the host),in a manner that is robust to a variety of attacks aimed at destroying the watermark while preserving the usability of the host.In contrast,our objective here is to obtain techniques for em-bedding high volumes of data in images,in a manner that causes minimal perceptual degradation,and is robust to“benign”JPEG compression attacks.The latter would occur naturally,for exam-ple,when an image with embedded data is being transmitted over a link with limited capacity.

Information-theoretic prescriptions for data hiding typically focus on hiding in independent and identically distributed(i.i.d) Gaussian host samples.Roughly speaking,these guidelines trans-late to hiding the data by means of the choice of quantizer for the host data,as?rst observed by Costa[1],based on results of Gel’fand and Pinsker[2]on coding with side information(with the host data playing the role of side information).Moulin et.al.

[3]have subsequently built on these results to provide a game-theoretic analysis of data hiding,with the hider and attacker as ad-versaries,and have provided parallel Gaussian models for images that facilitate application of information-theoretic analyses.Chen and Wornell[4]present a variety of practical approaches based on This research was supported in part by a grant from ONR#N0014-01-1-0380.similar ideas.A scalar quantization based data hiding scheme,to-gether with turbo coding to protect the hidden data,is considered in[5],while a trellis coded vector quantization scheme is consid-ered by Ramchandran et.al.[6].

In this paper,we consider scalar quantization based hiding schemes(as in[4])in the Discrete Cosine Transform(DCT)do-main.According to information-theoretic prescriptions for paral-lel Gaussian models of images in the DCT domain[3],data should be hidden in the low and mid frequency ranges of the host image, which have larger energies.However,we?nd that,in order to hide high volumes of data with low perceptual distortion,it is essential to add local perceptual criteria regarding which host coef?cients to hide data in.Two different schemes for making such decisions are provided.By optimizing for JPEG attacks,even without coding,it is possible to attain practical hiding capacities that are comparable to the capacity estimates given in[3]for AWGN attacks.

2.QUANTIZATION BASED DATA HIDING

2.1.Embedding data in choice of quantizer

As in[4],signature data symbols can be embedded in the host medium through the choice of scalar quantizer.For example,con-sider a uniform quantizer of step size,used on the host’s coef-?cients in some transform domain.Let odd reconstruction points represent a signature data bit‘1’.Likewise,even multiples of are used to embed‘0’.Thus,depending on the bit value to be em-bedded,one of two uniform quantizers of step size is chosen. Moreover,the quantizers can be pseudo-randomly dithered,where the chosen quantizers are shifted by a pseudo-random sequence available only to encoder and decoder.As such,the embedding scheme is not readily decipherable to a third party observer,with-out explicit knowledge of the dither sequence.

Decoding is performed by quantizing the received coef?cient to the nearest reconstruction point of all quantizers.An even re-construction point indicates that a‘0’has been hidden.Likewise, if a reconstruction point lies on an odd quantizer,a‘1’has been hidden.

2.2.JPEG attacks

JPEG compression is a common and practical attack model for data embedded in digital images,and arises naturally when com-pression is used to?t the modi?ed image into communication links or memories of varying capacities.JPEG operates on88blocks of DCT coef?cients.The coef?cient,,of each block is quantized uniformly with step size,taken from an88JPEG

quantization matrix.The JPEG quantization matrix is determined by the level of desired compression,or quality factor(QF).Quality factors range from0to100,100corresponding to no compression,

and75being a typical amount of compression.

It is well understood that high frequency distortion in images is less perceptible than its low frequency equivalent.Accordingly, JPEG uses?ner quantizers for low frequency coef?cients.De-pending on the quality factor,most mid to high coef?cients are

quantized so coarsely that their reconstruction value is zero.The quantized coef?cients are subsequently run-length and entropy en-coded.

For a DCT domain scalar quantization embedding scheme to survive such an attack,the spacing between a‘0’and‘1’quan-tizer must be at least.This can be guaranteed by adopting the

JPEG quantization matrix and using odd multiples of to em-bed a‘1’and even multiples of to embed a‘0’,when hiding in coef?cient.Thus,one can tune a DCT coef?cient quantization embedding scheme to guarantee survival of the signature data for

a given amount of JPEG compression.In fact,data hidden in this fashion will be robust to all JPEG compression attacks lesser than or equal to that of the design quality factor.

2.3.Performance penalty under A WGN attack

While our scalar quantization based scheme is well matched to

JPEG compression attacks,it does incur a substantial penalty for the worst-case AWGN attack.We quantify this in the context of an i.i.d.Gaussian host as follows.Letting and denote the mean squared embedding induced distortion and mean squared at-

tack distortion,the hiding capacity with AWGN attack is given by[1,7],in the small regime that typical data hiding systems operate.We compare this“vector ca-pacity”(termed thus because the optimal strategy involves vector

quantization of the host)to that of a scalar quantizer embedding scheme with hard decision decoding.Letting denote the dis-tance between a‘0’and‘1’quantizer,the variance of the quanti-zation error is approximately.The probability of bit error is given by

(1)

where denotes the complementary cumulative distribution func-tion of a standard Gaussian random variable.The capacity of a bi-nary symmetric channel with transition probability is given by [8],where

.

As with the vector capacity,the scalar capacity is solely a func-tion of,the distortion to noise ratio(DNR).Figure1shows roughly a loss due to the suboptimal encoding strategy used here,a gap that can be closed using soft decisions and vector quan-tization.

3.PERCEPTUAL EMBEDDING CRITERIA:TWO

APPROACHES

In the previous section we described how coef?cients are quan-tized to embed information bits.Next we decide which coef?-cients should be used for embedding.This will have a signi?cant effect on the perceptual quality of the embedded image.We have devised two such approaches–(i)entropy thresholding and(ii)

Fig.1.Gap between scalar and vector quantizer data hiding sys-tems.

selectively embedding in coef?cients.Both use some criterion to decide which coef?cients should be used to embed so that the per-ceptual quality of the host image is preserved.Thus,the amount of data hidden is adapted to the characteristics of the host.

3.1.Entropy thresholding

For a quantizer hiding scheme based on JPEG,one observes less distortion when embedding in low frequency DCT coef?cients be-cause of JPEG’s?ner quantization in this range.However,JPEG uses predictive encoding for the DC coef?cients,,of succes-sive blocks,so the additive uniform noise model does not apply. Furthermore,distortion induced in these coef?cients would not be localized to their88block.We therefore do not use these to embed data.

Next,we computed the2-norm entropy of each88block as follows,

(2) Only those coef?cient blocks whose entropy exceeds a predeter-mined threshold are used to hide data.Likewise,the decoder checks the entropy of each88block to decide if data has been hidden.The threshold entropy is determined by the desired em-bedding rate(or allowable distortion)for a particular image.

In general,compression will decrease the entropy of the coef?-cient blocks.Therefore,it is necessary to check that the entropy of each block used to embed information,compressed to the design quality factor,still exceeds the threshold entropy.If a particular block fails this test,we keep it as such,and embed the same data in the next block that passes the test

3.2.Selectively embedding in coef?cients

The above thresholding scheme uses an entropy criterion to deter-mine when to embed in a DCT block.We can take this idea one step further.Instead of embedding in a?xed number of coef?cients in qualifying blocks,we now decide to embed information on a co-ef?cient by coef?cient basis.In this manner,we embed precisely in those coef?cients that cause minimal perceptual distortion.

Coef?cients that are not quantized to zero by the design JPEG quantizer are used to embed information.Quantization embed-ding is performed as usual when embedding a‘1’.If a‘0’is to be embedded,the coef?cients are quantized to even reconstruction values.However,if this results in quantization to zero,we leave

it as such,and the same’0’is embedded in the next coef?cient satisfying the non-zero criterion.The decoder simply disregards all coef?cients that quantize to zero.Otherwise,decoding is per-formed as usual.

Selecting non-zero coef?cients for embedding minimizes the perceptual distortion incurred.Many image coef?cients are very close to zero once divided by the JPEG quantization matrix,and would be quantized to zero upon JPEG compression.To embed’1’in such coef?cients introduces a large amount of distortion relative to the original coef?cient size,a factor which seems to be percep-tually important.This is avoided by choosing not to use zeros for embedding.

4.RESULTS

4.1.Entropy thresholding

The entropy thresholding scheme was implemented to withstand JPEG compression at various quality factors.A“just noticeable”criterion for embedding induced distortion was used to determine the entropy threshold and number of low frequency coef?cients used per qualifying block with a512512Lena test image.Ta-ble1shows the compression(bits per pixel),number of embedded bits,distortion to noise ratio(DNR),and embedding rate results at each QF.Note,our and denote the mean squared em-bedding induced distortion and mean squared attack distortion,re-spectively.Figure2shows the compressed,hidden Lena image at50and75quality factors.The performance of this scheme is severely degraded at QF=25.Note that decoding of the embed-ded data is perfect for all JPEG attacks lesser than or equal to the design QF.

By con?ning attention to JPEG attacks,we are able to achieve large embedding rates without employing any error correction cod-ing.Our empirical results cannot be compared directly with the ca-pacity estimates in[3],since the latter were derived assuming that both the hider and the attacker use an optimal strategy(forming a so-called saddlepoint solution for the data hiding game considered there),whereas we use a suboptimal hiding strategy optimized for a suboptimal(JPEG compression)attack.In principle,therefore, our capacity can be smaller or larger than the estimates in[3].As it happens,for,1we are able to embed at a rate of 0.132bits per pixel in Lena against a JPEG attack,which is signi?-cantly larger than the corresponding saddlepoint capacity estimate of0.04bits per pixel in[3](which corresponds to optimal embed-ding for a worst-case AWGN attack).

As expected,when the JPEG attack is replaced by an AWGN attack inducing the same distortion,the performance of our schemes deteriorates.Figure3plots,for both the entropy thresholding and coef?cient based data hiding schemes,the bit error rate(BER)ver-sus DNR for an AWGN attack.The uncoded BER is signi?cant, which shows that,while our uncoded,hard-decision based,sys-tems are ideally matched to JPEG attacks,error correction coding must be employed in order to handle other additive attacks.In such a setting,we anticipate that achieving capacity will require use of more sophisticated vector quantization based schemes for embed-ding,as well as powerful codes with soft decision decoding.

1Here denotes the average mean squared error induced by the hider, and the average mean squared error induced by the attacker.The nota-tion differs from that in[3],where denotes the sum of the mean squared errors induced by both the hider and

attacker.

(a)0.66bpp

(QF=50)(b)1.04bpp(QF=75)

Fig.2.Entropy thresholding scheme

QF compression(bpp)#bits DNR(dB)rate(bpp) 250.424,970 2.90.019 500.6615,344 3.80.059 75 1.0434,460 6.50.132

Table1.Performance of thresholding scheme.

4.2.Selectively embedding in coef?cients

The coef?cient based data hiding scheme was implemented to with-stand JPEG compression at different quality factors.Table2has the size of the JPEG attacked composite image in bpp,total num-ber of embedded bits,DNR,and embedding rate at each QF for the Lena test image.Figure4shows the original and compressed composite Lena images for the various quality factors.Decoding is perfect for all JPEG attacks lesser than or equal to the design QF.

The coef?cient scheme operates in the high DNR regime be-cause it does not use an entropy criterion to discriminate between 88DCT blocks and actually does most of JPEG’s work,thus al-locating minimal power to the attack channel.Figure3shows the BER for an AWGN attack with as given in the QF=75imple-mentation.Again,channel codes would be used to deal with the degraded performance under an AWGN attack,with a correspond-ing loss in rate.

5.CONCLUSIONS

The key contribution of this paper is the use of perceptual crite-ria for embedding in images,in conjunction with the quantization based embedding prescribed by information theory in the context of simple host models.Both of our embedding methods are highly optimized for JPEG compression attacks,which enables them to offer excellent performance without error correction coding.In particular,the capacity obtained without coding under JPEG at-tack using our schemes is comparable to the capacity estimates QF compression(bpp)#bits DNR(dB)rate(bpp) 250.3811,04526.20.042 500.6018,73022.50.071 750.9429,87119.00.114

Table2.Performance of coef?cient scheme.

Fig.3.BER for AWGN attack

under AWGN attacks in[3],attaining which in general would pre-sumably require complex embedding and decoding schemes.

It is relatively straightforward to add standard error correction (including the use of soft decisions and iterative decoding[5])to make our methods robust to non-JPEG attacks,such as the AWGN attack or wavelet compression.Another extension is the use of vector quantization techniques such as those in[6].Our use of perceptual criteria to determine the embedding locations(rather than only statistical criteria which may specify which frequency bands to embed in),however,result in a new requirement in terms of error correction coding.Since the decoder must now decide which blocks or coef?cients the data is hidden in,our methods are potentially vulnerable to insertion and deletion errors.Thus,an important direction for future work is the incorporation of insertion and deletion codes[9],in addition to standard error correction into our hiding schemes.

Our coef?cient embedding scheme demonstrates superior per-formance at high JPEG compression.This scheme minimizes the power allocated to the JPEG attack channel,thereby operating in a high DNR regime.Another key advantage of this scheme is that it adapts the amount of data embedded to the characteristics of the host image(i.e the rate of this scheme is limited by the num-ber of non-zero scaled DCT coef?cients).However,for certain highly textured images,in which we were able to embed at very high rates(bits per image),the decoder suffered from insertion and deletion errors of about.

The entropy thresholding scheme allows the data hider greater ?exibility between embedding rate and induced distortion.On av-erage,compression will lower the entropy of the embedded image. In a few cases,the entropy of a coef?cient block is actually in-creased,causing the decoder to look for data in unused blocks. We have observed less than inserted/deleted blocks in the test images.However,this problem is exacerbated when the attack quality factor is mismatched to that of the encoder.

While hiding based on the JPEG quantization matrix was a convenient choice for illustrating our ideas,the performance of our hiding schemes can be improved by using a more powerful com-

pression mechanism.For

example,hiding data in the wavelet

do-

main promises to

be a robust scheme that would survive a wavelet

compression attack by design,and likely survive JPEG compres-sion with small error rates.Thus,applying our results to this do-main is an avenue for future work.

(a)Original Lena(b)0.94bpp(QF=75)

(c)0.59bpp(QF=50)(d)0.38bpp(QF=25)

Fig.4.Coef?cient based scheme

6.REFERENCES

[1]M.H.M.Costa,“Writing on dirty paper,”IEEE Transactions

on Information Theory,vol.29,no.3,pp.439–441,May1983.

[2]S.I.Gel’Fand and M.S.Pinsker,“Coding for channel with

random parameters,”Problems of Control and Information Theory,vol.9,no.1,pp.19–31,Jan.1979.

[3]P.Moulin and M.K.Mihcak,“The data-hiding capacity of

image sources,”Preprint,June2001.

[4] B.Chen and G.W.Wornell,“Quantization index modulation:

A class of provably good methods for digital watermarking

and information embedding,”IEEE Transaction on Informa-tion Theory,vol.47,no.4,pp.1423–1443,May2001.

[5]M.Kesal,M.K.Mihcak,R.Koetter,and P.Moulin,“Iter-

atively decodable codes for watermarking applications,”in Proc.2nd Int.Symp.on Turbo Codes and Related Topics,Sept.

2000.

[6]J.Chou,S.S.Pradhan,and K.Ramchandran,“A robust opti-

mization solution to the data hiding problem using distributed source coding principles,”in Proceedings of Conference on Information Sciences and Systems(CISS),Mar.2000.

[7]P.Moulin and J.A.O’Sullivan,“Information-theoretic analy-

sis of information hiding,”Preprint,Jan.2001.

[8]T.M.Cover and J.A.Thomas,Elements of Information The-

ory,Wiley,1991.

[9]M.C.Davey and D.J.C.Mackay,“Reliable communication

over channels with insertions,deletions,and substitutions,”

IEEE Transactions on Information Theory,vol.47,no.2,pp.

687–698,Feb.2001.

UIUC汽车检测图像数据库(UIUC Image Database for Car Detection)

UIUC汽车检测图像数据库(UIUC Image Database for Car Detection) 数据介绍： This database contains images of side views of cars for use in evaluating object detection algorithms. 关键词： 汽车,探测,目标检测算法,目标,算法, cars,car,detection,object detection algorithms,object,algorithms,algorithm, 数据格式： TEXT 数据详细介绍： UIUC Image Database for Car Detection ●Description This database contains images of side views of cars for use in evaluating object detection algorithms. The images were collected at UIUC by Shivani Agarwal, Aatif Awan and Dan Roth, and were used in the experiments reported in [1], [2]. ●Data information The download package contains the following: 1050 training images (550 car and 500 non-car images) 170 single-scale test images, containing 200 cars at roughly the same scale as in the training images 108 multi-scale test images, containing 139 cars at various scales

英国文科类专业申请的情况

免费澳洲、英国、新西兰留学咨询与办理 官网：https://www.sodocs.net/doc/011992951.html, 英国文科类专业申请的情况 随着2019年英国申请季的开始，选专业又成为我们面临的重大事情。今天我们主要帮助学生梳理一下英国文科类专业申请的情况。 英国文科类的专业主要包括：教育学、政治学、社会学和人类学、传媒等。教育学 顾名思义就是研究当老师的学问。英国大学教育学专业分支丰富，不仅有倾向于教学的分支，例如倾向于教学方法的分支， 主要培养教学这个方向。还有倾向于管理的分支。例如教育领导管理的分支。主要培养学校的行政管理人员。所以如果想去大学当辅导员的学生，可以考虑这个专业分支哦。 政治学

免费澳洲、英国、新西兰留学咨询与办理 官网：https://www.sodocs.net/doc/011992951.html, 顾名思义研究国家和国际政治的专业。英国大学政治学主要开设专业分支有政治理论、国际关系和公共政策等。这一类专业申请比较多的是国际关系，因为国际关系相对于政治学，学习内容更加具体。例如国际关系会关注国际安全、人权与公平正义、比较政治经济学等。申请该专业的优势在于不需要专业背景，一般接受转专业申请的学生。分数要求也不高，例如去年有一个学生来自福建师范大学，本科是传播学，82分。申请到曼彻斯特大学和伯明翰大学。对于条件比较普通 的学生，可以考虑这个专业。 近期有一个学生来自于上海师范大学天华学院，本科国际商务贸易专业，分数是77分，学生比较想去好学校，在推荐学校和专业时，首先从文科出发，相对于教育学和传媒，学生申请国际关系更能申请到比较到的学校。 传媒 传媒属于我们申请的热门专业之一。传媒主要包括新闻，电影，媒体和创意产业等专业。新闻专业要求比较高的写作水平，所以新闻专业不太好申请，除非写作功底比较好的同学可以尝试。电影专业分支比较适合本科专业就是电影专业，因为课程涉及到一些动画设计等课程。媒体类和创意产业属于大家选择比较多的分支，因为专业背景比较宽泛，雅思要求比较适中，一般都是总分要求6.5（6.0）。典型学校有利兹大学、诺丁汉大学、华威大学、格拉斯哥大学、谢菲尔德大学。如果条件比较适中的学生可以选择纽卡斯尔大学、莱斯特大学、东英吉利亚大学等。去年有一个三本的学生，均分为 83，本科就读汉语言文学专业，拿到了上述3个大学的offer 。 人类学和社会学

运筹学课程设计指导书

运筹学课程设计指导书 一、课程设计目的 1、初步掌握运筹学知识在管理问题中应用的基本方法与步骤； 2、巩固和加深对所学运筹学理论知识及方法的理解与掌握； 3、锻炼从管理实践中发掘、提炼问题，分析问题，选择建立运筹学模型，利用模型求解问题，并对问题的解进行分析与评价的综合应用能力； 4、通过利用运筹学计算机软件求解模型的操作，掌握运筹学计算软件的基本操作方法，并了解计算机在运筹学中的应用； 二、课程设计内容与步骤 第一部分是基本实验，为必做部分；需要每位同学单独完成，并写出相应的实验报告。第二部分是提高部分，题目自选或自拟，锻炼综合应用运筹学知识及软件解决实际问题的能力；可以单独完成，也可以合作完成（最多3人一组），写出相应的报告。 1、基本实验在完成基本实验后，每位同学要按照实验要求完成实验报告，实验报告应包括问题描述、建模、上机求解、结果分析及答辩几方面。实验报告必须是打印稿（word文档等），手写稿无效。请大家按照要求认真完成实验报告，如果两份实验报告雷同，或相差很少，则两份实验报告均为0分，其它抄袭情况，将根据抄袭多少扣分。（约占总分的70%） 2、提高部分根据自己的兴趣或所查找的资料，从实际情况出发，自拟题目；在实验报告中，陈述问题，建立模型，求解，结果分析，此部分应着重突出自己的观点和想法。（此部分按照排名先后给分，约占总分的30%） 三、课程设计要求 1、实验目的 学会建立相应的运筹学模型 学会Excel、Lindo和WinQSB，QM for windows软件的基本使用方法 学会用Excel、Lindo和WinQSB，QM for windows软件得到问题的最优解 2、实验要求 分析问题、建立模型,并阐明建立模型的过程； 说明并显示软件使用和计算的详细过程与结果； 结果分析,将结果返回到实际问题进行分析、评价。 四、题目内容 （一）Excel规划求解基本实验 1、雅致家具厂生产4种小型家具，由于该四种家具具有不同的大小、形状、重量和风格，所以它们所需要的主要原料（木材和玻璃）、制作时间、最大销售量与利润均不相同。该厂每天可提供的木材、玻璃和工人劳动时间分别为600单位、1000单位与400小时，详细的数据资料见下表。问： （1）应如何安排这四种家具的日产量，使得该厂的日利润最大？ （2）家具厂是否愿意出10元的加班费，让某工人加班1小时？ （3）如果可提供的工人劳动时间变为398小时，该厂的日利润有何变化？ （4）该厂应优先考虑购买何种资源？

Hybrid images

Copyright ? 2006 by the Association for Computing Machinery, Inc. Permission to make digital or hard c opies of part or all of this work for personal or classroom use is granted without fee provided that copies are not made or distributed for c ommercial advantage and that c opies bear this notic e and the full citation on the first page. Copyrights for components of this work owned by others than ACM must be honored. Abstracting with credit is permitted. To copy otherwise, to republish, to post on servers, or to redistribute to lists, requires prior spec ific permission and/or a fee. Request permissions from Permissions Dept, ACM Inc., fax +1 (212) 869-0481 or e-mail permissions@https://www.sodocs.net/doc/011992951.html, . ? 2006 ACM 0730-0301/06/0700- $5.00 0527Hybrid images Aude Oliva ?MIT-BCS Antonio Torralba ?MIT-CSAIL Philippe.G.Schyns ?University of Glasgow Figure 1:A hybrid image is a picture that combines the low-spatial frequencies of one picture with the high spatial frequencies of another picture producing an image with an interpretation that changes with viewing distance.In this ?gure,the people may appear sad,up close,but step back a few meters and look at the expressions again. Abstract We present hybrid images ,a technique that produces static images with two interpretations,which change as a function of viewing distance.Hybrid images are based on the multiscale processing of images by the human visual system and are motivated by masking studies in visual perception.These images can be used to create compelling displays in which the image appears to change as the viewing distance changes.We show that by taking into account perceptual grouping mechanisms it is possible to build compelling hybrid images with stable percepts at each distance.We show ex-amples in which hybrid images are used to create textures that be-come visible only when seen up-close,to generate facial expres-sions whose interpretation changes with viewing distance,and to visualize changes over time within a single picture.Keywords:Hybrid images,human perception,scale space 1Introduction Here we exploit the multiscale perceptual mechanisms of human vi-sion to create visual illusions (hybrid images )where two different interpretations of a picture can be perceived by changing the view-ing distance or the presentation time.We use and extend the method originally proposed by Schyns and Oliva [1994;1997;1999].Fig.1shows an example of a hybrid image assembled from two images ?e-mail:oliva@https://www.sodocs.net/doc/011992951.html, ?e-mail:torralba@https://www.sodocs.net/doc/011992951.html, ?e-mail: p.schyns@https://www.sodocs.net/doc/011992951.html, in which the faces displayed different emotions.High spatial fre-quencies correspond to faces with ”sad”expressions.Low spatial frequencies correspond to the same faces with ”happy”and ”sur-prise”emotions (i.e.,the emotions are,from left to right:happy,surprise,happy and happy).To switch from one interpretation to the other one can step away a few meters from the picture.Artists have effectively employed low spatial frequency manipu-lation to elicit a percept that changes when relying on peripheral vision (e.g.,[Livingstone 2000;Dali 1996]).Inspired by this work,Setlur and Gooch [2004]propose a technique that creates facial im-ages with con?icting emotional states at different spatial frequen-cies.The images produce subtle expression variations with gaze changes.In this paper,we demonstrate the effectiveness of hybrid images in creating images with two very different possible interpre-tations. Hybrid images are generated by superimposing two images at two different spatial scales:the low-spatial scale is obtained by ?ltering one image with a low-pass ?lter;the high spatial scale is obtained by ?ltering a second image with a high-pass ?lter.The ?nal im-age is composed by adding these two ?ltered images.Note that hybrid images are a different technique than picture mosaics [Sil-vers 1997].Picture mosaics have two interpretations:a local one (which is given by the content of each of the pictures that compose the mosaic)and a global one (which is best seen at some prede?ned distance).Hybrid images,however,contain two coherent global image interpretations,one of which is of the low spatial frequen-cies,the other of high spatial frequencies. We illustrate this technique with several proof-of-concept exam-ples.We show how this technique can be applied to create face pictures that change expression with viewing distance,to display two con?gurations of a scene in a single picture,and to present tex-tures that disappear when viewed at a distance. 2The design of hybrid images A hybrid image (H )is obtained by combining two images (I 1and I 2),one ?ltered with a low-pass ?lter (G 1)and the second one ?l-

MATLAB与在运筹学中的应用

MATLAB与在运筹学中的应用 摘要：论文通过MATLAB在运筹学中的应用实例，探讨了MATLAB在运筹学中的应用方法和技巧，初步了解matlab中优化工具箱的使用。 关键字：MATLAB应用运筹学优化计算 引言 运筹学是近代应用数学的一个分支，主要是研究如何将生产、管理等事件中出现的运筹问题加以提炼，然后利用数学方法进行解决的学科。运筹学是应用数学和形式科学的跨领域研究，利用像是统计学、数学模型和算法等方法，去寻找复杂问题中的最佳或近似最佳的解答。运筹学经常用于解决现实生活中的复杂问题，特别是改善或优化现有系统的效率。运筹学中常用的运算工具有Matlab、Mathematica、Maple、SAS 、SPSS、Lindo/Lingo、GAMS、WinQSB、Excel、其他，如SQP、DPS、ORS、Visual Decision、Decision Explore、AIMMS、Crystal等。 Matlab是矩阵实验室（Matrix Laboratory）的简称，是美国MathWorks公司出品的商业数学软件，和Mathematica、Maple并称为三大数学软件。 用于算法开发、数据可视化、数据分析以及数值计算的高级技术计算语言和交互式环境，主要包括Matlab和Simulink两大部分。 主要应用于工程计算、控制设计、信号处理与通讯、图像处理、信号检测、金融建模设计与分析等领域。 将matlab用于运筹学的最优化运算可以很好的解决优化问题，而且matlab 还专门有优化工具箱，是处理优化问题更加方便。 一、例：0-1规划（《运筹学》80页例3-9） 求minZ=x1-3*x2+6*x3+2*x4-4*x5 6*x1+2*x2-x3+7*x4+x5<=12 约束条件 x1+4*x2+5*x3-x4+3*x5>=10 Xj=0或1，j=1,2,3,4

运用线性规划对运输问题研究

运用线性规划对运输问题研究 班级：金融103班姓名：王纬福学号：5400210132摘要：由于企业选择运输路线或运输工具不合理而导致物流运输成本不能最小化的问题普遍存在而管理运筹学却能很好的解决此问题。通过科学的方法对问题进行具体化再建立数学模型并求解，就能找到运输成本最小的运输组合。 关键词：物流运输成本、输成本、管理运筹学、WinQSB2.0、线性规划 一、引言 日常生活中，人们经常需要将某些物品由一个空间位置移动到另一个空间位置，这就产生了运输。如何判定科学的运输方案，使运输所需的总费用最少，就是管理运筹学在运输问题上的运用需要解决的问题。 运输问题是一类应用广泛的特殊的线性规划问题，在线性规划的一般理论和单纯形法出现以前，康托洛维奇（L.V.Kant）和希奇柯克（F.L.Hitchcock）已经研究了运输问题。所以，运输问题又有“康-希问题”之称。对于运输问题（Transportation Problem TP）当然可用前面所讲的单纯形法求解，但由于该问题本身的特殊性，我们可以找到比标准单纯形法更简单有效的专门方法，从而节约计算时间和费用。主要是因为它们的约束方程组的系数矩阵具有特殊结构，使得这类问题的求解方法比常规的单纯形法要更为简便。 一、研究现状 运输问题的研究较多，并且几乎所有的线性规划书中都有论述。遗憾的是一些书中所建立的数学模型都不够全面和系统的。但是也有一些模型是严谨的没有漏洞和缺陷，并且很容易在此基础上修改或添加一些其他约束条件便于在实际工程中进行应用。管理运筹学在运输问题上的研究较为深入、全面、系统。对于计算机软件的引用也很前言，winQSB2.0对于普通甚至深入研究运输问题就已经是简单而又使用、耐用、好用的了。现在相关的杂志、期刊都越来越多关于管理运筹学，关于运输问题的文章论文初版，越来越得到重视。 二、文献回顾 随着物流行业和企业对物流运输要求的不断提高，企业的面临着更大的市场竞争，其运输活动在企业不断发展过程中，面临着越来越大难度的运输组合的选择决策问题。如何正确解决这个问题，是企业能够持续经营和发展不可忽视和必须面对的。这个问题同时也引起了企业界、学术界等社会各界的广泛关注。运输问题的实质是企业与运输组合的经济性问题，成功的企业通常都会面临如何选取最佳运输组合或运输路线这样一个重要问题，即以企业运输成本最小化作为确定最佳运输组合或运输路线的原落脚点。 四、案例分析 例：某公司下设生产同类产品的加工厂A1、A2、A3，生产的产品由4个销售点B1、B2、B3、B4出售。各工厂的生产量、各销售点的销量以及各工厂到各销售点的单位运价如下表：

武汉工程大学文科基金项目

所属学科及学科代码： 项目编号： 武汉工程大学文科基金项目 申请书 项目名称： 项目负责人： 联系电话： 依托学院部门： 申请日期： 武汉工程大学科技处制 2007年9月

简表填写要求 一、简表内容将输入计算机，必须认真填写，采用国家公布的标准简化汉 字。简表中学科（专业）代码按GB/T13745-92“学科分类与代码”表填写。 二、部分栏目填写要求： 项目名称——应确切反映研究内容，最多不超过25个汉字（包括标点符号）。 学科名称——申请项目所属的第二级或三级学科。 申请金额——以万元为单位，用阿拉伯数字表示，注意小数点。 起止年月——起始时间从申请的次年元月算起。 项目组其他主要成员——指在项目组内对学术思想、技术路线的制定理论分析及对项目的完成起主要作用的人员。

一、项目信息简表

二、选题：本课题国内外研究现状述评；选题的意义。 三、内容：本课题研究的基本思路和方法；主要观点。 四、预期价值：本课题理论创新程度或实际应用价值。 五、研究基础：课题负责人已有相关成果；主要参考文献。 六、完成项目的条件和保证：包括申请者和项目组主要成员业务简历、项目申请人和主要成员承担过的科研课题以及发表的论文；科研成果的社会评价；完成本课题的研究能力和时间保证；资料设备；科研手段。 （请分5部分逐项填写）。

七、经费预算

六、项目负责人承诺 我确认本申请书及附件内容真实、准确。如果获得资助，我将严格按照学校有关项目管理办法的规定，认真履行项目负责人职责，积极组织开展研究工作，合理安排研究经费，按时报送有关材料并接受检查。若申请书失实或在项目执行过程中违反有关科研项目管理办法规定，本人将承担全部责任。 负责人签字： 年月日 七、所在学院意见 负责人签字：学院盖章： 年月日 八、科技处审核 已经按照项目申报要求对项目申请人的资格及项目申请书内容进行了审核。项目如获资助，科技处将根据项目申请书内容，落实项目研究所需经费及其它条件；以保证项目按时顺利完成。 科技处盖章 年月日

202X美国留学文科专业申请建议.doc

202X美国留学文科专业申请建议 现在，去美国留学，文科专业比较容易申请一些，但是，美国的文科专业众多，该如何申请呢，下面来说说美国留学文科专业申请建议。 1、文科专业非常庞杂，常见专业有：语言类，新闻和传播，政治学，社会学，人类学，历史学，经济学，法学，教育学，心理学，建筑学，城市规划和景观设计，艺术类等。 2、若是英语专业，可以申请教育学，比如教育心理，TESOL，早期教育等；文学类，比如比较文学等；传媒类，公关，广告等；或者政治学、社会学、历史学。具体申请什么专业根据申请人的背景经历进行确定。中文和日语专业的可以申请东亚研究，有东亚研究设置的专业都是TOP50的学校，所以申请难度也是很大的。 3、美国的顶尖大学综合排名前30的学校里设置传播学院的并不多。因为传播学是后兴起的专业才有七、八十年的历史，很多老牌的学校他们排斥这样的新兴学科，所以他们不设置这样的学院，例如哈佛，耶鲁，牛津，剑桥等根本没有传播学校。有些学校虽然有传播学院或者有相关的专业但是规模一般也很小，有些学生甚至只招生本校的本科生。例如：斯坦福大学，传播学一直没有权威的专业排名，比较出名的学校有：哥伦比亚大学，纽约大学，但是这两个牛校都只设置新闻学院，并没有传播学研究。而且这两个学校的新闻类专业是属于那种超级牛人才能申请到的，一般在国内每年招生只有1-2个，甚至没有。总的来说新闻类专业申请难度很大。并且对申请人的英语及GRE成绩要求非常高，并且申请人最好有在央视，新华社，这样的背景才会比较有利。 4、政治学，社会学，人类学和历史，一般硕士的申请很难拿到奖学金，博士奖学金设置会很多，但是同时对学生的研究兴趣有要求，并希望看到学生对未来的职业发展规划。

Mosaicing of acoustic camera images

Mosaicing of acoustic camera images K.Kim,N.Neretti and N.Intrator Abstract:An algorithm for image registration and mosaicing on underwater sonar image sequences characterised by a high noise level,inhomogeneous illumination and low frame rate is presented.Imaging geometry of acoustic cameras is signi?cantly different from that of pinhole cameras.For a planar surface viewed through a pinhole camera undergoing translational and rotational motion,registration can be obtained via a projective transformation.For an acoustic camera,it is shown that,under the same conditions,an af?ne transformation is a good approximation.A novel image fusion method,which maximises the signal-to-noise ratio of the mosaic image is proposed.The full procedure includes illumination correction,feature based transformation estimation,and image fusion for mosaicing. 1Introduction The acquisition of underwater images is performed in noisy environments with low visibility.For optical images in those environments,often natural light is not available, and even if arti?cial light is applied,the visible range is limited. For this reason,sonar systems are widely used to obtain images of seabed or other underwater objects. An acoustic camera is a novel device that can produce a real time underwater image sequence.Detailed imaging methods of acoustic cameras can be found in[1].Acoustic cameras provide extremely high resolution(for a sonar)and rapid refresh rates[1].Despite those merits of acoustic cameras over other sonar systems,it still has shortcomings compared to normal optical cameras: (i)Limitation of sight range:Unlike optical cameras which have a2-D array of photosensors,acoustic cameras have a 1-D transducer array.2-D representation is obtained from the temporal sequence of the transducer array.For this reason,it can collect information from a limited range. (ii)Low signal-to-noise ratio(SNR):The size of the transducers is comparable to the wavelength of ultrasonic waves,so the intensity of a pixel depends not only on the amplitude,but also on the phase difference of the re?ected signal.This is the reason for the Rician distribution of the ultrasound image noise.In addition,there is often a background ultrasound noise in underwater environments. It follows that the SNR is signi?cantly lower than in optical images. (iii)Low resolution with respect to optical images:owing to the limitation in the transducer size,the number of transducers that can be packed in an array is physically restricted,and so is the number of pixels in the horizontal axis.For example,a mine reacquisition and identi?cation sonar(MIRIS)has64transducers[1].(iv)Inhomogeneous insoni?cation:The unique geometry of an acoustic camera requires the sonar device to be aligned parallel to the surface of interest,so that the whole surface falls within the vertical?eld of view of the acoustic camera [1].This alignment is not always trivial,and the misalign-ment often makes dark areas in acoustic camera images. The above limitations can be addressed by image mosai-cing,which is broadly used to build a wider view image [2–4],or to estimate the motion of a vehicle[5,6].For ordinary images,mosaicing is also used for image enhancement such as denoising,deblurring,or super-resolution[7,8]. There has been extensive research on image mosaicing, and its applications[9–13].However,standard methods for image registration[14,15]are not directly applicable to acoustic camera images,because of the discrepancy of image quality,inhomogeneous insoni?cation pro?le,and different geometry.Marks et al.have described a mosaicing algorithm of the ocean?oor taken with an optical camera [2].Rzhanov et al.have also described a mosaicing algorithm of underwater optical images resulting in high resolution seabed maps[3].Both of them deal with a similar problem of illumination,but use different methods:image matching by edge detection and Fourier based matching, which are not directly related to our work.In addition,since their mosaicing algorithms are not intended for image quality enhancement,we need to come up with a different mosaicing algorithm. In this paper,we describe a mosaicing algorithm for a sequence of acoustic camera images.We show that an af?ne transformation is appropriate for images taken from an acoustic camera undergoing translational and rotational motion.We propose a method to register acoustic camera images from a video sequence using a feature matching algorithm.Based on the parameters of image registration,a mosaic image is built.During the mosaicing,the image quality is enhanced in terms of SNR and resolution. 2Properties of acoustic camera images Sonar image acquisition includes several steps,insoni?ca-tion,scattering,and detection of the returning signal.In this Section,we describe physical aspects of images acquired from acoustic lens sonar systems,or acoustic cameras. q IEE,2005 IEE Proceedings online no.20045015 doi:10.1049/ip-rsn:20045015 The authors are with the Institute for Brain and Neural Systems,Brown University,Box1843Providence RI02912,USA E-mail:kio@https://www.sodocs.net/doc/011992951.html, Paper?rst received21st May2004and in revised form22nd April2005

高中理科申请书

篇一：高一文理分科申请表 高一文理分科申请表 明： 1、传 媒含播音与主持艺术、广播电视编导、服饰艺术与表演、影视表演、空中乘务等专业；美术含 绘画、书法艺术和书法教育、设计等专业；音乐含声乐、舞蹈、器乐、理论作曲、指挥等专业。 2、本 申请表经学生、家长、班主任签名后，不能再更改，学校以此为依据重新分班。 上梅 中学教务处 2013年12月28日 篇二：文理分科申请书 请书 敬的老师： 我 是，现在高（）班，本人喜欢科技制作、科普读物、科技发明和生命科学等相关 知识与能力锻炼，对理科的兴趣大于对文科的兴趣。经过慎重考虑，并与家长商量，我决定申 请就读（文科、理科、美术、体育、音乐、舞蹈、传媒）班，敬请批准。 谢谢！ （学 生本人签名）年月日 意上述申请。 （家 长签名）年月日 篇三：转班申请书 转班申请书尊敬的杨老师： 您好！ 我想转到理科高二（16）班.因为经过假期的思考，我发现自己对文科没有很大的兴趣，而且

我也了解到自己在文科方面很难得到提高。对此，我恳请杨老师能让我转到理科高二（16） 班，我真切希望自己能转理科高二（16）班，我总结了我转班的原因，有以下几点： 一、 我觉得自己对文科的兴趣没有了之前的那种热情，而且我觉得自己在理科方面还有待提高，我 对它也很感兴趣。人们都说兴趣是学习最好的老师，有了兴趣就是成功的一半。而现在我对文 科已经没有了那种热情，又怎么会对学习文科尽心尽力、认真努力的去做呢？所以我真心的想 转到理科。 二、 我的文科成绩不是很好，而且从小我就贪玩把英语落下了很多。然而英语在文科当中可以占很 大的优势，我的英语很差，读文科没有优势，所以我希望自己能转到理科。 三、 学习文科需要很好的记忆力，但我这个人很赖，不喜欢背诵，而文科又需要背诵和 忆。 四、 根据社会的需求和我个人的发展空间，我想理科更适合我。 亡羊 补牢，为时不晚，希望杨老师能给我一次机会。 此致 敬礼 申请 人： 011年7月31日 篇四：申请书 尊敬的政府领导： 我叫 xxx，19xx年xx月xx日生，系xxx居民，配偶19xx年生，也是xxx人，我们都无固定 职业，现有家庭成员x人，家庭月收入xxx元。我们也曾经多次想买房子，但就我们这点收 入想买完全属于我们自己的房子那简直是奢望。所以本人和家人至今也一直居无定所，但为了 生活，又不得不在城区内四处奔波打工，并租房居住。由于本人家庭生活的实际困难和无住房 的实际情况，现想申请政府廉租住房一套，望领导给予批准为盼！ 谢谢！ 请人：

机器学习_CMU Face Images Data Set(CMU人脸图像数据集)

CMU Face Images Data Set(CMU人脸图像数据集) 数据摘要： This data consists of 640 black and white face images of people taken with varying pose (straight, left, right, up), expression (neutral, happy, sad, angry), eyes (wearing sunglasses or not), and size 中文关键词： 人脸,图像,分类,UCI, 英文关键词： Faces,Image,Classification,UCI, 数据格式： TEXT 数据用途： This data set is used for classification. 数据详细介绍： CMU Face Images Data Set

Abstract: This data consists of 640 black and white face images of people taken with varying pose (straight, left, right, up), expression (neutral, happy, sad, angry), eyes (wearing sunglasses or not), and size Source: Original Owner and Donor: Tom Mitchell School of Computer Science Carnegie Mellon University tom.mitchell '@' https://www.sodocs.net/doc/011992951.html, https://www.sodocs.net/doc/011992951.html,/~tom/ Data Set Information: Each image can be characterized by the pose, expression, eyes, and size. There are 32 images for each person capturing every combination of features. To view the images, you can use the program xv. The image data can be found in /faces. This directory contains 20 subdirectories, one for each person, named by userid. Each of these directories contains several different face images of the same person. You will be interested in the images with the following naming convention: .pgm is the user id of the person in the image, and this field has 20 values: an2i, at33, boland, bpm, ch4f, cheyer, choon, danieln, glickman, karyadi, kawamura, kk49, megak, mitchell, night, phoebe, saavik, steffi, sz24, and tammo. is the head position of the person, and this field has 4 values: straight, left, right, up. is the facial expression of the person, and this field has 4 values: