Computers in Industry 62 (2011) 32–41

Physimetric identification (Physi-ID)—Applying biometric concept in physicalobject identification

S.K. Kwok *, Ocean P.H. Ng, Albert H.C. Tsang, H.M. Liem

Department of Industrial and Systems Engineering, The Hong Kong Polytechnic University, Hung Hom, Kowloon, Hong Kong, China

A R T I C L E I N F O

Article history:

Received 17 April 2009

Received in revised form 14 April 2010

Accepted 31 May 2010

Available online 3 July 2010

Keywords:

Physimetric identification

Biometrics

Automatic identification

Physical object authentication

Physical asset management

Anti-counterfeit

A B S T R A C T

By relying on business intelligence technologies, services can be delivered to customers automatically by

computer systems. To provide the right services to the right person, a methodology that precisely

identifies a personal’s identity must be in place. Biometrics offers a secure and reliable method for

computerized personal identification and authentication. It accurately recognizes and determines the

unique identity of a person based on her physiological and/or behavioral characteristics. In the case of

physical objects, they may also be required to be identified automatically in order to provide additional

services, such as at checkout counters of supermarkets or customs clearance checkpoints. Numerous

crimes and business losses (e.g. counterfeit products) are related to mis-identification of physical

objects. This paper introduces physimetric identification, an approach that applies the concept of

biometrics for physical object identification. It addresses the problem through authenticating physical

objects based on their unique physical and/or chemical characteristics. Apart from introducing the

concept of physimetric identification, issues such as real applications, deployment considerations and

limitations of the proposed technology will also be discussed.

� 2010 Elsevier B.V. All rights reserved.

Contents lists available at ScienceDirect

Computers in Industry

journa l homepage: www.e lsevier .com/ locate /compind

1. Introduction

Identifying physical objects is an essential aspect of industrialapplications. Physical objects must be identified through the use ofdifferent technologies ranging from pen and paper, barcode, 2Dbarcode, to Radio Frequency Identification (RFID), so that they canbe managed and operated. In the current IT era, through the use ofsuitable identification technologies, computer systems can recog-nize physical objects automatically by reading the barcodes orRFID tags attached to these objects. By combining artificialintelligent and decision support technologies, such systems canautomatically identify physical objects and take appropriateactions accordingly. There are, however, drawbacks associatedwith existing technologies. One of the major limitations is thatcomputer systems recognize physical objects by verifying theidentification technology used on the object instead of by verifyingthe physical object itself. Mis-identification problems will exist inindustrial applications if mistakes or errors are made by theautomatic identification technologies used.

It is realized that identification technology applied to humanbeings is well developed and mature. Biometrics is widely adopted

* Corresponding author. Tel.: +852 2766 6578; fax: +852 2362 9308.

E-mail address: mfskkwok@inet.polyu.edu.hk (S.K. Kwok).

0166-3615/$ – see front matter � 2010 Elsevier B.V. All rights reserved.

doi:10.1016/j.compind.2010.05.014

in different areas to provide effective and accurate identification. Itfollows that we can improve physical object identification byexploiting the lessons learnt from the success of humanidentification technology. Every person has distinctive physiologi-cal and behavioral characteristics (e.g. face, voice). Our brains areable to verify these characteristics and match them with patternsin our memories. This way, humans can recognize each otheraccordingly. The use of biometrics enables computers to accuratelyidentify a person by verifying the person’s physiological and/orbehavioral characteristics since the mid 19th century. It is notedthat physical objects also contain some specific physical orchemical characteristics which can be used for unique identifica-tion. If those characteristics such as texture pattern and spectra canbe captured, measured and verified by the computer system, andintegrated with existing identification technologies, the accuracyof physical object identification can be improved significantly.

This paper introduces the concept and mechanism of Physi-metric identification (Physi-ID) which will improve the accuracy inidentification of physical objects by verifying their unique physicaland chemical properties. In the following sections, the features andlimitations of current identification technologies will be discussed.Biometrics as a mature human identification technology will thenbe reviewed. The workflow and advantages of the proposed Physi-ID will then be explained in detail. At the end of this paper,application examples of Physi-ID will be presented, along with

S.K. Kwok et al. / Computers in Industry 62 (2011) 32–41 33

discussion on the deployment considerations and limitations ofthe technology.

2. Review on identification and authentication technologies

2.1. Current automatic identification technologies

Automatic identification (Auto-ID) is developed to deliverobject information to computer systems that support businessoperations [1]. It acts as an ID container that stores the digitalidentity of an object. Tagged objects are identified by the process ofautomated extraction of digital identity stored in the ID container.The extracted information is processed by a computer system forfurther manipulation. Nowadays, nearly all products are packagedwith a barcode for its identification. In recent years, RFID becomesanother hot Auto-ID technology for physical object identificationin various industries. The value and benefits of using Auto-IDtechnology to facilitate automatic physical asset managementhave been reported in operations such as retail management [2],manufacturing management [3] and supply chain management[4]. Auto-ID typically involves the use of a serial number, i.e., an ID,which is related to the physical object. Details of an object can beretrieved from the database by using the object’s serial number(ID) as the key. Therefore if an RFID tag on a bottle of water is readby an RFID reader, the system captures the information stored inthe tag, and then uses the information to retrieve data from thedatabase to identify the physical object as a bottle of water. Thisapproach changes the traditional way of physical object identifi-cation, and can be performed without human involvement. Itdramatically speeds up item management operations and signifi-cantly reduces human effort and error in data entry. The captureddata can be automatically fed into enterprise applications forfurther processing.

The introduction of new Auto-ID technologies, including 2Dbarcode and RFID, and the mature development of Internet andnetworking infrastructure make unique object identificationbecome possible. Unique identification defined by Departmentof Defence (DoD) of the United States Government refers to a set ofdata representing tangible assets that are globally unique andunambiguous. Integrity and quality of the unique ID must beassured throughout the life cycle of the object being identified, soas to support multi-faceted business applications and users [5].Unique identification aims to enhance the visibility of physicalobjects in the supply chain in order to improve item managementand accountability, and finally to achieve product life cycleinformation management. To achieve unique identification, apartfrom embedding globally unique product identifier (e.g. RFID) inthe product, there must be a linking mechanism to productinformation that may be stored in backend systems and a networkapproach to share unique product information among variousparties around the world. Distributed information architectures forcollaborative Logistics (DIALOG), World Wide Article Information(WWAI) and Electronic Product Code (EPC) are the currentlyknown approaches to implement unique identification [6]. In theexample of EPC and EPCglobal network infrastructure proposed byAuto-ID Center is designed to achieve seamless sharing of RFID-related data. Products with unique ID (EPC) are traced and trackedfrom end to end of the supply chain, and information of theproducts is shared among parties in this network [7,8]. In thepharmaceutical industry, unique object identification is beingadopted to trace and track each item of the pharmaceuticalproduct. With such technology and network, the product pedigreewhich is a certified record that contains information about eachdistribution of prescription drug [9] can be retrieved for theenhancement of logistics visibility, product authentication andanti-counterfeit applications [10,11].

2.2. Limitations of current automatic identification technologies

While Auto-ID technology enables total automation of uniquephysical asset management and availability of real-time data inenterprise applications, it is not faultless. Many research studiesshow the unreliability of Auto-ID applications and systems. Whenno corrective action is taken, even a low data error rate may createsevere problems and data inaccuracy in computer system [12]. Theunderlying reason is that the computer system identifies physicalobjects by making reference to the information stored in theattached Auto-ID instead of the object itself. The major problem ofthis approach is that the system will deliver erroneous data whenmistakes are made in the Auto-ID reading process, such ascapturing incorrect or non-existent data known as reading ghosttags, and failing to capture existent data (e.g. missing tags) in RFIDsystems [13].

The existing applications of Auto-ID technology assume that theinformation stored in the Auto-ID is always precise and the datacapture process is error-free. In practice, these assumptions are ofteninvalid, and existing automatic physical object identificationapplications typically do not have exception handling routines tocope with those situations. Human involvement is thus required inthese cases to deal with reading errors of the system, limiting therealization of the full benefits of automatic identification. At retailstores, human operators are still needed to present the goods tobarcode scanners to eliminate reading errors. More serious problemswill happen when a genuine Auto-ID is cloned and the cloned one isattached to a counterfeit product. This is a serious weakness of anti-counterfeit and authentication applications. Barcode scan is anexample that has this limitation of Auto-ID. Visualize this: adishonest shopper used a home-use printer to produce the UPC(Universal Product Code) barcodes for a lower priced product, andthen placed them over the correct bar codes on higher pricedproducts. Since products are identified purely on the basis of thebarcode read by the Point-of-Sale system, the dishonest shopper willbe able to buy higher priced products at lower prices undetected.Wal-Mart lost USD 1.5 million in 2004 because of this fraud [14].

Readability is always a challenge of automatic physical objectidentification. In the case of barcode, it will be unreadable whenblocked from the line of sight of the barcode reader, or when the laserbeam of the barcode reader does not target the barcode accurately. Inthe case of RFID, readability is influenced by radio wave reflection,refraction, interference, RFID tag orientation, and other environ-mental factors [15]. Lodewijks et al. [16] concluded that reliability ofRFID systems depends on the application factors, the RF technologyused and deployment environment. In addition, environmentalchange may distort system reliability. Thus, there is a need for theverification of the physical object itself in automatic identification ofphysical objects. Reliability of the current Auto-ID applications is farfrom the Six Sigma level. The three major types of errors made byAuto-ID systems are: (1) physical object with improper Auto-ID, (2)physical object without Auto-ID and (3) Auto-ID without physicalobject. All these will generate inaccurate information to themanagement system and thus trigger inappropriate actions.

3. Lessons learnt from human identification technology(Biometrics)

In pursuit of a better approach to physical object identification,lessons may be learnt from Biometrics – the relatively maturehuman identification technology. Biometrics is also one of theautomatic identification technologies commonly applied toverification of living individuals by using physiological andbehavioral characteristics [17,18]. A biometric system essentiallyincludes a pattern recognition system which acquires biometricdata from an individual, a feature extraction module or program

Table 1Mechanisms of person identification and authentication.

Types of identification and authentication mechanisms Person identification Object identification

Type I: Authentication by Knowledge (Something They Know) Passwords, PIN (Personal Identification Numbers), etc. Not applicable

Type II: Authentication by Ownership (Something They Have) Keys, Auto-ID technologies Conventional Auto-ID technologies

Type III: Authentication by Characteristic (Something They Are) Fingerprints, DNA patterns, etc. Raman Spectrum, Texture pattern, etc.

Fig. 1. Biometric identification and physimetric identification.

S.K. Kwok et al. / Computers in Industry 62 (2011) 32–4134

which extracts a useful feature set from the acquired data, and acomparison mechanism to determine the similarity between thefeature set and the template set in the database [19]. Typicalexamples of biometrics are DNA, fingerprint, face, iris, voice andsignature.

The advantage of biometrics is verifying and identifying thetarget object by inherent distinctive characteristics. It is the mostsecure and accurate identification mechanism. Mechanisms foridentification and authentication of person can be classified intothree types, as summarized in Table 1 [20]. Type I is authenticationby knowledge, type II is authentication by ownership and type III isauthentication by characteristic. Type III authentication mecha-nism is most reliable because it does not have the problem of theftand cloning, which is common to Type I and Type II mechanisms. Inthe case of personal identification, the characteristics recognizedby Type III mechanisms are those unique ones that are very closelytied to a particular person.

A physical object, however, is very different from a person in anidentification process. Type I approach is not applicable because aphysical object cannot provide knowledge to authenticate itsidentity. Type II approach has been implemented by usingconventional Auto-ID technologies and its shortcomings has beenexplained above. A practical solution to implement Type IIImechanism is yet to be developed. Although most physical objectsare produced in large quantities under the same manufacturingcondition, some characteristics such as Raman Spectrum [21] canbe used to distinguish one physical object from another.

A single application of Type III authentication is not secureenough. Examining multiple biometric characteristics (e.g. multi-method biometrics by verifying iris and fingerprint together) andintegrating data stored with multiple identification technologies(e.g. RFID and Smart Card), will provide a more secure and accuratemeans of authentication [22]. Electronic identification cards,electronic visas or e-passports are common examples that applythis form of Auto-ID and authentication technology. An electronicidentification card stores both the passport ID and fingerprint dataof the cardholder for verification [23]. Some countries may apply amulti-method biometric system to verify fingerprint and facecharacteristics such as Mainland Travel Permit Card for Hong Kongand Macau Residents.

It can be reasoned that one or more physical/chemicalcharacteristics of a physical object can be examined in theauthentication process, and this process can be integrated withAuto-ID technologies. As a result, an approach to Auto-IDverification that is based on the unique characteristics of aphysical object is designed. This can significantly improve accuracyof the authentication process.

4. Physimetric identification

Physimetric identification (Physi-ID) is an authenticationtechnology that integrates the features of Types II and IIImechanisms – authentication by ownership and characteristics– in order to make authentication more reliable and accurate. Itverifies the identity of a physical object not only by Auto-ID butalso by the physical features of that object (Fig. 1). Type IIauthentication by ownership is achieved by verifying the Auto-ID.Type III authentication by characteristic is done by verifying the

physimetric features of the object. The word ‘Physimetric’combines ‘Physi’, representing physical object, and ‘metric’,representing a measurement. Consequently, a strong relationshipbetween a physical object and its Auto-ID can be established. WithPhysi-ID, an application system can distinguish a bottle of waterfrom a can of soda, even if a cloned tag for a bottle of water isattached to a can of soda.

Physi-ID, therefore, holds both the object’s ID and data of itsphysical features for automatic verification. The ID represents theidentity of the physical object, and the data of physical feature isused for verifying the Physi-ID’s ownership and the object’sidentity. The physical feature used for verifying the object is namedas the physimetric feature in the Physi-ID. It can be any type ofphysical properties such as Raman spectrum, texture feature,dimensions, weight, colour pattern, shape, and light reflective/refractive index. The design of Physi-ID helps to address thelimitations of current Auto-ID technologies identified in Section2.2.

To achieve in situ authentication without network support, thedata of physimetric feature can be stored in the memory of Auto-ID. It provides the most convenience way of instant authentication.And this approach is suitable for use in applications while networkis not available. Conversely, the physimetric data can be stored inthe backend database and associated with the Auto-ID whennetwork is ready for data communication with backend system.This approach is more suitable for applications which large volumeof physimetric data needs to be processed.

The concept of Physi-ID is presented in Fig. 2. The alert-freesituation takes place when the object is attached with a properAuto-ID. If the object is attached with an improper Auto-ID, theconventional Auto-ID system will create a wrong record of theobject according to the information captured from the Auto-ID andthe mistake may not be discovered. In the case of Physi-ID, an alertwill be generated since the physimetric feature stored in the Auto-ID does not match with that of the physical object. The second type[(Fig._1)TD$FIG]

[(Fig._2)TD$FIG]

Fig. 2. Concept of Physi-ID.

S.K. Kwok et al. / Computers in Industry 62 (2011) 32–41 35

of common problematic situation occurs when the Auto-ID tag isdetached from its physical object and that Auto-ID is read by thesystem. Traditionally, the system will mistakenly record theexistence of a physical object, but Physi-ID can detect the missingphysical object and generate an alert. In another scenario, when aphysical object without an attached Auto-ID tag is presented foridentification, the system will accurately recognize the situationrather than reporting that no object has been detected. Table 2summarizes the comparison of identification results of traditionalphysical object identification and Physi-ID. The Physi-ID approacheffectively eliminates incorrect identification, and provides alertsfor various types of identification problems.

Fig. 3 illustrates the workflow of a Physi-ID applicationimplemented with RFID technology without network supported.After the measurement of a physimetric feature, data of thephysimetric feature and the unique ID are programmed into anRFID tag, which is then attached to the associated object for futureidentification. During the verification process, the object is placedin the read range of an RFID reader to retrieve the stored data. Inthe meantime, the specified physimetric feature of the object isalso measured by an electronic measurement device. Thecomputer system compares the data of the physimetric featureretrieved from the RFID tag with the actual physimetric featuremeasured on the spot. If the two match with each other, the objectis identified and authenticated. If not, an identification alert will begenerated. Details of the mechanism for the verification ofphysimetric properties involving the use of Raman Spectrum areshown in Fig. 4. First of all, the spectral data of the physical object iscaptured by a Raman spectroscope. The captured data istransformed and feature extracted to become a sample set data.After controlling the quality of the sample set data, the processeddata is verified by a pattern recognition algorithm (spectral pattern

Table 2Comparison between the identification results of Physi-ID and traditional physical obj

Identification result

Traditional physical ob

Identification result

Proper Auto-ID attached to physical object Identified

Improper Auto-ID attached to physical object Identified

Auto-ID only without being attached to physical object Identified

Physical object only without Auto-ID attached No response

analysis) that compares it with the template set data retrievedfrom the Auto-ID. Decisions are made according to the similarityvalue computed by the algorithm. In case the data does not meetthe required quality standard, the sample collection process will berepeated until a qualified sample set is obtained.

The feature extraction and pattern matching algorithms aredeveloped according to the specific characteristics of physimetricdata. Since physimetric identification is based on the verification ofthe stored physimetric properties (template set) with theproperties measured from the physical object (sample set), ithas the same limitation as biometrics that there may be matchingerrors (False Accepts and False Reject) caused by imperfect sampleset collection, and by changes of properties and environmentalconditions [19]. All these factors are evaluated and considered inthe development of the physimetric identification algorithm. Theverification is normally done by applying a similarity function S

(XA, XA1) to compare the extracted sample set data XA withtemplate set data XA1. If the value of S exceeds the definedthreshold value t, the identity is affirmed; otherwise, the identity isrejected. In using different physimetric properties, the similarityfunction and threshold are to be determined from the trade offbetween the false match rate (FMR) and the false nonmatch rate(FNMR) for specific applications. In addition, multi-methodphysimetric identification mechanism can also be applied in orderto improve accuracy of identification.

5. Applications of physimetric identification

In the previous sections, problems of Auto-ID applications havebeen discussed and Physi-ID is proposed as a solution. BasicallyAuto-ID faces two critical problems: inaccurate reading and theproblem of cloning. Physi-ID is designed to address these

ect identification methods.

ject identification Physi-ID

Remark Identification result Remark

Correct identification Identified Correct identification

Incorrect identification Identified and alert Correct identification

Incorrect identification Identified and alert Correct identification

Incorrect identification Identified and alert Correct identification

[(Fig._3)TD$FIG]

Fig. 3. Workflow of Physi-ID with RFID technology.

S.K. Kwok et al. / Computers in Industry 62 (2011) 32–4136

problems. In this section, two possible applications of Physi-ID,namely, quantity and quality control, and anti-counterfeit system,are discussed.

5.1. Physi-ID in quantity and quality control

Auto-ID is widely used as an enabling technology of automaticquantity control. It is used in factories, warehouses, and retailstores to keep track of quantities of physical objects. Data accuracyand level of automation are two main concerns in this type ofapplications. These issues arise from the limitations of currentAuto-ID technologies. Human involvement is always required in anAuto-ID tracking workflow to assure reading quality; this is costlyand inefficient. The findings of a study on a postal RFID application[24] highlight this problem. In that study, a number of experimentswere conducted on parcels on pallet to test whether all the taggedparcels are identified accurately. The results show that it is difficultto achieve 100% recognition. The problem becomes one of themajor obstacles to applying Auto-ID in postal applications. Physi-ID presented in this paper is a preferred mechanism to ensureaccuracy of identification. It prevents inaccurate readings from

[(Fig._4)TD$FIG]

Fig. 4. Physimetric id

being entered into the computer system and provides alerts tohuman operators to isolate the problematic object(s). For example,an RFID gateway can be integrated with an electronic scale forPhysi-ID to check the weight of the physical object as one of itsphysimetric features. When a forklift truck carrying taggedphysical objects passes through the RFID gateway and electronicscale, the system can determine the total weight of the load bysumming up the weights of individual items retrieved from theRFID tags of the physical objects, and compare it with the weightmeasured by the electronic scale. Even though, physical objectswith RFID can be invisible without line of sight, the system can stillidentify the present of invisible physical objects (by measuring thetotal weight). It ensures that no missing and ghost reading, orincorrect data will be entered into the computer system (see Fig. 5).This mechanism can also be similarly applied in postal RFIDapplications by using the parcel’s weight as physimetric feature toprevent erroneous readings and thus enhance the reliability ofsuch systems.

Physi-ID can be further extended as an inspection mechanismin the process of quality control. Certain physimetric featuresstored in the Auto-ID can serve as the quality requirements of the

entification flow.

[(Fig._5)TD$FIG]

Fig. 5. Physi-ID in a quantity checking process.

S.K. Kwok et al. / Computers in Industry 62 (2011) 32–41 37

goods. If the physimetric feature reaches the required or acceptedrange of quality requirement, the quality of that physical object isverified. The quality checking process that applies to products withhigh precision surface can be an example of this type of application[25]. For those products, surface roughness is measured by asurface profile measurement machine and a surface roughnessprofile is generated in a digital format (see Fig. 6). By applyingPhysi-ID, the acceptable range of surface roughness profile can beencoded into Auto-IDs and attached to the products. During thequality checking process of a tagged product, an Auto-ID reader canbe used to extract information on product ID and physimetricfeature data (surface roughness profile) from the attached Auto-ID.The product’s surface roughness is measured by the measurement

[(Fig._6)TD$FIG]

Fig. 6. Surface profile measurement machine – Wyko NT8000 optical profiling system (Ph

machine. This way, the surface quality of each item can be verifiedindividually and automatically.

5.2. Physi-ID in anti-counterfeit applications

Cloning and copying have always been problems that frustratethe effectiveness of anti-counterfeit applications. Many cases haveshown that Auto-ID can be copied easily. In 2006, Lukas Grunwald,a consultant with computer security firm DN-Systems, demon-strated cloning of RFID-enabled German passports [26]. However,cloning and copying will not be a high risk problem in biometrics-based personal identification, because it involves the verification ofunique personal characteristics. As pointed out by Randy

oto source: VEECO Metrology Group) and digitized data of surface roughness profile.

S.K. Kwok et al. / Computers in Industry 62 (2011) 32–4138

Vanderhoof, executive director of Smart Card Alliance, it would notbe different from stealing someone else’s passport and trying topresent that as one’s own at a border entry point. The samesituation, unfortunately, may create huge problem to physicalobject identification since existing mechanisms generally do notrecognize the object’s physical characteristics. Physi-ID effectivelyaddresses the problems by strongly associating the ID informationwith the physical features. In anti-counterfeit applications, Physi-ID can successfully identify fake products even though they areattached with authentic Auto-ID tags.

In anti-counterfeit applications, the selection of physicalproperties with high discrimination power is very critical. Ramanspectroscopy is one type of characterization technique which canprobe the material property down to the molecular level,determining the vibrational modes of a particular material/objectsurface. The spectrum obtained can be used to infer the molecularformula and its orientation, and accordingly, is unique to aparticular material being measured. It may be easier to understandand has been widely accepted that the Raman spectroscopy itselfcan be used to identify different materials in different forms, suchas liquid, solid, and even gas. Raman spectroscopy is a welldeveloped optical characterization tool which offers advantagesover other experimental techniques for this specific task, includingthe fact that it is noninvasive, it has considerable, i.e., about 1 mm,spatial resolution, and suitable types of lasers can be selected toavoid fluorescence. It has been used to monitor the surfacemolecular orientation of polymer solution [27], solid film surface[28,29], dynamics of hemoglobin [30,31], thermal phase transition[32,33], in textile industry [34,35], and even in the interactionbetween cellulose fibers and external deformation by monitoringthe stress-induced specific Raman band shifts [36–38]. Recently,this technique is applied to determine the contents of pharma-ceutical capsules [39], cocaine dissolved in beverages [40], andconcealed liquid within bottles and packaging [41]. Since genuineproduct has distinctive Raman spectral data when compared tothat of fake ones, the spectrum encoded in Physi-ID will act as aunique material fingerprint, ‘‘a molecular signature’’, for Physi-IDbased counterfeit prevention procedures.

A measurement of Raman spectrum on two visually identicalleather products – a real leather product and an imitated leatherproduct – was made with results shown in Fig. 7. Somecharacteristic peaks are observed in one leather product butnot in the other. This indicates that the real leather item possesses

[(Fig._7)TD$FIG]

Fig. 7. (a) An image of a ‘‘real’’ wallet, (b) an image of an ‘‘imitated’’ wallet. The crosses ind

spectra taken from these two visually identical wallets. (For interpretation of the refere

article.)

a different type of vibrational mode contributed by some types ofmolecules that could not be found in the imitated leather. Thus, itcan be concluded that two leather products are of two differentmaterials. The unique features can be extracted and used incombination with Auto-ID technologies such as RFID forautomatic authentication of products. To further enhance thesecurity of Physi-ID anti-counterfeit solution, multiple physi-metric properties on different spots of the object may be used andthe data protected by cryptographic secure key functions (e.g.AES–Advanced Encryption Standard and DES–Data EncryptionStandard) and hash functions (e.g. MD5–Message-Digest algo-rithm 5 and SHA-1–Secure Hash Algorithm), to prevent unautho-rized retrieval of the physimetric information for fake productmanufacturing.

5.3. Implementation issues of Physi-ID

To implement Physi-ID, a number of considerations have to betaken into account in selecting the physimetric feature for specificapplications. Take biometrics as an example, these considerationsare universality, uniqueness, permanence and collectability [42].Universality means everyone should have this characteristic.Uniqueness means that the characteristic should be unique in theworld. Permanence means that the characteristic remains stable inthe object’s life cycle. Collectibility implies that the characteristiccan be measured and captured easily.

Since most physical objects are products for sale which shouldnot be damaged during the identification process, noninvasivenessis one of the most important selection criteria of physimetricproperties. Other considerations may differ from one application toanother. For quantity checking, or physical asset managementapplications, Physi-ID is used to ensure accuracy in tracking ofphysical objects. For such applications, the physimetric feature tobe selected is not necessary to be a unique feature. High degree ofuniversality, permanence and collectibility, however, are impor-tant considerations. Weight of products can be used as a suitablefeature for quantity checking since it meets the universality,permanence and collectibility requirements. In the case of qualitychecking, the objective is to ensure that the object meets an agreedquality standard. Consequently, whether the feature is easy to bemeasured and captured by electronic devices becomes the mostcritical factor. In anti-counterfeit applications, the physimetricfeature serves to be a factor that distinguishes itself from fake

icate the position of the laser excitation used in the measurement and (c) the Raman

nces to colour in this figure legend, the reader is referred to the web version of the

Table 3Selection criteria of physimetric feature in three types of application.

Degree of importance

Quantity

checking

Quality

checking

Anti-counterfeit

Noninvasiveness High High High

Uniqueness Low Low High

Universality High Low Low

Permanence High Low High

Collectibility High High Medium

Association with

the object’s value

Low Low High

Table 4Hardware cost of Physi-ID applications.

Physi-ID applications with RFID Hardware Cost

Quantity control (Section 5.1) (i) Electronic floor scale USD 2000

(ii) RFID reader and antenna USD 3000

(ii) Tags (unit price) USD 0.1

Anti-counterfeit (Section 5.2) (i) Small-scale Raman Spectrum USD 65000

(ii) RFID reader and antenna USD 3000

(iii) Tags (unit price) USD 0.1

[(Fig._8)TD$FIG]

Fig. 8. RFID reader and antenna set up in Midas Printing Group Company.

S.K. Kwok et al. / Computers in Industry 62 (2011) 32–41 39

products. Thus, in such cases, uniqueness and permanence are themost critical selection criteria. Besides the four factors mentionedabove, the association between the physimetric feature and thevalue of the object can be another consideration. This suggeststhat it is better to select a feature that reflects the object’s value,such as the refractive index of diamond. Table 3 summarizes theselection criteria of physimetric feature in different types ofapplication.

Cost is another critical consideration on the implementationof Physi-ID. The first investment must be the cost on electronicmeasurement devices and the integration with Auto-ID technol-ogies. This cost can be ranging from several hundred to severaltens of thousands US dollars depending on the applications andthe measured physimetric features. In addition, since the welldeveloped Auto-ID technology, the market available RFID and 2Dbarcodes already provide large memory size for storage ofphysimetric feature without high investment. Key features arestored in RFID and other attributes can be stored in backenddatabase by leveraging the network communication as dis-cussed. Table 4 lists the hardware cost of some Physi-IDapplications:

[(Fig._9)TD$FIG]

Fig. 9. Integration of electronic floor scale and RFID gateway for phy

A real-life implementation can be proposed to a printingcompany, Midas Printing Group Company Limited, which is a listedcompany with solid reputation in printing industry. An RFID-basedPhysical Asset Management System (Fig. 8 set up of RFID gatewayin warehouse & docking bay) has been already implemented in thiscompany [43], i.e. RFID is applied at pallet level and all items on thepallet are associated with the pallet RFID in the backend database.Tracking the pallet RFID is equal to trace the whole pallet of items.Unfortunately, sometimes, items are misplaced (or missing) on thepallet while packing or during transfer by different reasons. It maycause mis-delivery of goods. Same as other printing company,Midas also installed industrial grade electronic floor scale. To applyPhysi-ID, Midas only needs to integrate the floor scales with theRFID readers at selected checkpoints, such as warehouse exits/docking bays (see Fig. 9). The Physi-ID stored in the pallet RFIDconsists of Auto-ID and the total weight of associated items(derived from the packing list) on the pallet. Each time the palletpasses through the checkpoint, the integrity of packing is validatedand mis-delivery can be minimized.

It is summarized that the technology factor of physimetricidentification implementation is ready, and it is the time toeducate the various industries on the benefit of using physimetricidentification, and that research is continued on physimetricproperties to provide more and more cost effective solutions andapplications.

simetric quantity checking in Midas Printing Group Company.

S.K. Kwok et al. / Computers in Industry 62 (2011) 32–4140

6. Conclusions

This paper reviews the mechanisms and features of Auto-IDtechnologies in physical object identification. It points out that twomajor problems, namely inaccurate reading and cloning of Auto-ID, are faced by industry. Physi-ID is proposed as a new mechanismto address these problems by strongly associating the Auto-IDinformation with the physical characteristics of a particular object.The possible applications of the proposed system in quantitychecking, quality assurance and counterfeit detection are dis-cussed. Different applications require different considerations inthe selection of physimetric features. The criteria for the selectionof the physimetric feature to be used in the applications discussedare also specified.

In a similar manner to the first introduction of Biometrictechnology, the Physi-ID presented in this paper contributes (i) anovel idea on integrating Auto-ID technology with physimetricmeasurement to improve the current automatic identificationpractice and achieve a more reliable and feasible model ofautomatic physical object identification, (ii) initiating the researchand development on different physimetric measurement fordifferent applications (e.g. using Raman Spectrum for anti-counterfeit solutions). One of the future research directions ofPhysi-ID could focus on the technological aspects of thisapplication, such as improvement on the capabilities of theAuto-ID reader to enable simultaneous acquisition of ID informa-tion and physimetric feature. It is believed that Physi-ID willbenefit a number of industries, such as logistics and healthcare, inwhich process control and counterfeit detection are importantissues.

Acknowledgements

The authors would like to express their sincere thanks to theResearch Committee of The Hong Kong Polytechnic University forthe financial support of the research work presented in this paper.

References

[1] D. McFarlane, Y. Sheffi, The impact of automatic identification on supply chainoperations, International Journal of Logistics Management 14 (1) (2003) 1–17.

[2] P. Kourouthanassis, G. Roussos, Developing consumer-friendly pervasive retailsystems, Pervasive Computing IEEE 2 (2) (2003) 32–39.

[3] S.J. Udoka, P. Sanjiv Sarin, L.L. Massay, Applications of automatic identificationtechnologies across the I.E. curriculum, Computers & Industrial Engineering 29(1–4) (1995) 125–129.

[4] K. Kenny, W. Wu, S.K. Kwok, Design of an radio frequency identification basedintra supply chain system in the textile industry, Journal of Industrial Manage-ment and Data Systems 109 (9) (2009) 1166–1178.

[5] Item Unique Identification 101 The Basics, Department of Defense, USA, 2006.[6] K. Framling, M. Harrison, J. Brusey, J. Petrow, Requirements on unique identifiers

for managing product lifecycle information: comparison of alternativeapproaches, International Journal of Computer Integrated Manufacturing 20 (7)(2007) 715–726.

[7] EPCglobal standard, EPCglobal Tag Data Standards Version 1.3.1, GS1 EPCglobalInc., 2006.

[8] EPCglobal standard, Object Naming Service (ONS) Version 1.0, GS1 EPCglobal Inc.,2005.

[9] EPCglobal standard, Pedigree Ratified Standard Version 1.0, GS1 EPCglobal Inc.,2007.

[10] S.K. Kwok, A.H.C. Tsang, J.S.L. Ting, W.B. Lee, B.C.F. Cheung, An intelligent RFID-based electronic anti-counterfeit system (InRECS) for the manufacturing industry,in: Proceedings of the 17th International Federation of Automatic Control (IFAC)World Congress 2008, Seoul, Korea, July 6–11, 2008, (2008), pp. 5482–5487.

[11] S.K. Kwok, J.S.L. Ting, A.H.C. Tsang, W.B. Lee, B.C.F. Cheung, Design and develop-ment of a mobile EPC-RFID based self-validation system (MESS) for productauthentication, Computers In Industry, in press.

[12] K. Yun, S.B. Gershwin, Information inaccuracy in inventory systems: stock lossand stockout, IIE Transactions 37 (9) (2005) 843–859.

[13] K.M. Ramakrishnan, D.D. Deavours, Performance benchmarks for passive UHFRFID tags, in: Proceedings of the 13th GI/ITG Conference on Measurement,Modeling, and Evaluation of Computer and Communication Systems, 2006.

[14] Colin Fly, $1.5M Homemade Bar Code Scam, Colin Fly, The Associated Press, 2004December http://www.cbsnews.com/stories/2004/12/31/national/main664128.shtml.

[15] N.C. Wu, M.A. Nystrom, T.R. Lin, H.C. Yu, Challenges to global RFID adoption, in:Technology Management for the Global Future, PICMET 2006, vol. 2, 2006, 618–623.

[16] G. Lodewijks, H.P.M. Veeke, A.M.L.D.L. Cruz, Reliability of RFID in logistic systems,in: Proceedings of the IEEE International Conference, SOLI’06, 2006.

[17] J.L. Wayman, Fundamental of biometric authentication technologies, Internation-al Journal of Image and Graphics 1 (1) (2001) 93–113.

[18] A.K. Jain, R. Bolle, S. Pankanti, Introduction to Biometrics, Biometrics – PersonalIdentification in Networked Society, 1999, pp. 1–42.

[19] A.K. Jain, S. Prabhakar, An introduction to biometric recognition, IEEE Transac-tions on Circuits and Systems for Video Technology 14 (1) (2004) 4–20.

[20] P.R. Callagher, Guide to Understanding Identification and Authentication inTrusted Systems, National Computer Security Center, USA, 1991.

[21] J.C. Carter, W.E. Brewer, S.M. Angel, Raman spectroscopy for the in situ identifi-cation of cocaine and selected adulterants, Applied Spectroscopy 54 (12) (2000)1876–1881.

[22] D.T. Toledano, R.F. Pozo, A.H. Trapote, L.H. Gomez, Usability evaluation of multi-modal biometric verification systems, Interacting with Computers 18 (2006)1101–1122.

[23] N.K. Ratha, R. Bolle, Smartcard based authentication, in biometrics: personalidentification in networked society, The Kluwer Academic, International Series inEngineering and Computer Science 479 (18) (1999) 369–384.

[24] J-H. Park, J-H. Park, Postal RFID application model and performance, ETRI Journal28 (3) (2006) 405–408.

[25] M.N. Cheng, C.F. Cheung, W.B. Lee, S. To, A study of factors affecting surface qualityin ultra-precision raster milling, Key Engineering Materials 339 (2007) 400–406.

[26] M.C.O. Connor, Industry group says E-passport clone poses little risk, RFID Journal(2006) http://www.rfidjournal.com/article/articleprint/2559/-1/1/.

[27] H. Liem, P. Etchegoin, D.D.C. Bradley, Physical Review B 64 (2001) 1–5.[28] H. Liem, P. Etchegoin, K. Whitehead, D.D.C. Bradley, Journal of Applied Physics 92

(2) (2002) 1154–1161.[29] H-M. Liem, P. Etchegoin, K.S. Whitehead, D.D.C. Bradley, Advanced Functional

Materials 13 (2003) 66–72.[30] P. Etchegoin, H. Liem, R.C. Maher, L.F. Cohen, R.J.C. Brown, M.J.T. Milton, J.C. Gallop,

Chemical Physics Letter 367 (2002) 223–229.[31] P. Etchegoin, H. Liem, R.C. Maher, L.F. Cohen, R.J.C. Brown, H. Hartigan, M.J.T.

Milton, J.C. Gallop, Chemical Physics Letter 366 (2002) 115–121.[32] H. Liem, J. Cabanillas-Gonzalez, P. Etchegoin, D.D.C. Bradley, Journal of Physics:

Condensed Matter. 16 (2004) 721–728.[33] H. Liem, A natural length scale for the glass transition of conjugated polymer film,

Journal of Physics: Condensed Matter 19 (2007) 106–416.[34] H. Liem, L.Y. Yeung, J.L. Hu, Shape-memory polymers reduce the residual stresses

in cotton fibres, in: Research Highlights, Nature China, 2007.[35] H. Liem, L.Y. Yeung, J.L. Hu, Smart Materials & Structures 16 (2007) 748–753.[36] S.J. Eichhorn, J. Sirichaisit, R.J. Young, Deformation mechanisms in cellulose fibres,

paper and wood, Journal of Materials Science 36 (2001) 3129–3135.[37] W.Y Yeh, R.J. Young, Molecular deformation processes in aromatic high modulus

polymer fibres, Polymer 40 (4) (1999) 857–870.[38] S.J. Eichhorn, R.J. Young, The Youngs modulus of a microcrystalline cellulose,

Cellulose 8 (3) (2001) 197–207.[39] C. Eliasson, N.A. Macleod, L.C. Jayes, F.C. Clarke, S.V. Hammond, M.R. Smith, P.

Matousek, Non-invasive quantitative assessment of the content of pharmaceuti-cal capsules using transmission Raman spectroscopy, Journal of Pharmaceuticaland Biomedical Analysis 47 (2) (2008) 221–229.

[40] C. Eliasson, M. Claybourn, P. Matousek, Deep subsurface Raman spectroscopy ofturbid media by defocused collection system, Applied Spectroscopy 61 (10)(2007) 1123–1127.

[41] C. Ricci, C. Eliasson, N.A. Macleod, P.N. Newton, P. Matousek, S.G. Kazarian,Characterization of genuine and fake artesunate anti-malarial tablets using Four-ier transform infrared imaging and spatially offset Raman spectroscopy throughblister packs, Analytical and Bioanalytical chemistry 389 (5) (2007) 1525–1532.

[42] R. Clarke, Human identification in information systems: management challengesand public policy issues, Information Technology & People 7 (4) (1994) 6–37.

[43] S.K. Kwok, A.H.C. Tsang, B.C.F. Cheung, Realizing the Potential of RFID in Coun-terfeit Prevention, Physical Asset Management, and Business Applications: CaseStudies of Early Adopters, Department of Industrial and Systems Engineering, TheHong Kong Polytechnic University, 2007.

S.K. Kwok is a lecturer in the Department of Industrial

and Systems Engineering of The Hong Kong Polytechnic

University. His research areas are in artificial intelligence,

industrial and systems engineering, Information and

Communication Technologies (ICT), logistics enabling

technologies and mobile commerce. He participates in

several industrial-based research projects, which include

web-enabled collaborative working platform develop-

ment, customer relationship management, mobile

devices application in vendor management inventory,

RF-tag order tracking system, etc. For all the mentioned

projects, latest ICT are applied to smooth information

flow and enhance the Knowledge Management (KM)

among modern business units. The research outcomes are presented in several

international conferences and published in various international journals.

s in Industry 62 (2011) 32–41 41

P.H. Ng is an assistant technical officer at the

Department of Industrial and Systems Engineering of

The Hong Kong Polytechnic University. He received his

BSc degree and MEng degree in Industrial and Systems

Engineering at the same university. His research

interests include software engineering, logistics sys-

tem, and radio frequency identification (RFID) applica-

tions.

H.C. Tsang is the principal lecturer of the Department of

Industrial and Systems Engineering of The Hong Kong

Polytechnic University. He had provided consultancy

and advisory services to enterprises and industry

support organisations in manufacturing, logistics,

public utilities, healthcare and government sectors on

matters related to quality, reliability, maintenance,

performance management and assessment of perfor-

mance excellence. He is the author of ‘WeibullSoft’, a

computer-aided self learning package on Weibull

S.K. Kwok et al. / Computer

analysis and a co-author of two books: Reliability-Centred Maintenance: A Key

to Maintenance Excellence (published in 2000) and Maintenance, Reliability and

Replacement: Theory and Applications (published in 2006).

H.M. Liem is doing a postdoctoral stay in the Industrial

and Systems Engineering department at the Hong Kong

Polytechnic University to manage the PCB technology

centre, developing High-efficiency heat dissipation PCB

for High-Brightness LED. He completed his research

appointment in the Department of Textiles and

Clothing at the same university with focus being paid

to develop shape memory polymers for smart textile

application in 2006. He finished his first postdoctoral

research appointment in the industrialization of

organic thin film devices at the University of California

at Los Angeles in 2004. He earned his PhD degree in

Physics (material science and optics) at the Imperial

College London in year 2003. He has published 21 papers and has been awarded: (1)

Top papers 2004 in the Journal of Physics, Condensed Matter, Institute of Physics

Publishing, and (2) Research Highlights in 2007 in the Nature China, Nature

Publishing Group. He is also a regular referee of Institute of Physics, UK, and

America in Institute of Physics, and Nature China.