WO1996030217A1

WO1996030217A1 - Machine-readable label

Info

Publication number: WO1996030217A1
Application number: PCT/NZ1996/000021
Authority: WO
Inventors: Harold Terrence Salive; Michael David Barns-Graham
Original assignee: Kiwisoft Programs Limited
Priority date: 1995-03-31
Filing date: 1996-03-18
Publication date: 1996-10-03
Also published as: CA2215026A1; NO974285L; FI973635A; KR19980703503A; HUP9801931A3; EP0817727A1; JPH11502654A; FI973635A0; AU5127196A; TW318916B; NO974285D0; HUP9801931A2; ZA962295B; BR9607985A

Abstract

A label (100) for identification purposes, has a compact matrix of machine-readable cells (101-109) holding information and including the BCH error-correcting code. The cells may be binary, ternary (grey levels) or have more levels and may use colours. These labels are both easily detectable and accurately readable, because a plurality may occur within a single complex image captured outdoors, where adverse lighting and contaminants may detract from ideal conditions. Accordingly some (usually 4) characteristic cells (101, 102, 103, 104) per label are reserved for use as locators which, when found, indicate position, orientation and values on the scale of grey levels used in the symbol. A grey or white label perimeter may be used to determine intra-label lighting variations, as described in relation to the disclosed reading steps. One or more portable digital camera captures images and relays them to an analysis and decoding machine.

Description

MACHINE-READABLE LABEL

FIELD OF THE INVENTION

This invention relates to the field of identification, more particularly to identifying objects by means of labels, and most particularly to labels which are adapted for reading by machine (that is, a digital computer).

BACKGROUND

There is a need for an identifying label for use in tracking merchandise during shipping or manufacture. An optical or other type of "no-touch" label is preferred, and it should be capable of use in applications where:-

(a) image capture is carried out in possibly adverse circumstances, such as outdoors or where part of the label may be obscured, lost, or contaminated, curled up, or in shade, or the label may be presented obliquely or unsharply,

(b) the exact position and orientation of the label is not defined, (c) there may be a plurality or large number of labels in any one captured image,

(d) a high degree of accuracy in reading the label is required, and

(e) labels may be out of reach to a bar-code scanner or the like.

Furthermore, where, in particular, there is a need to use a cheap and relatively low-resolution camera every available part of the label should be covered with data markings that are as large as possible, given that a relatively small label such as one 50 mm across is highly preferred.

A particular application is in tracking cut tree trunks (lumber) from felling to export. At the forest site where logs are accumulated before loading onto a truck, it is customary in some forests to attach a pre-printed bar-coded label to the cut end of the trunk in order to identify the tree, which is separately described in relation to ownership, volume, quality, type, and the like. The bar-code and the associated data are later entered into a management computer database. A barcode is not easy to read later, when for example the log may be one of many on a moving truck, or in a cradle being loaded onto a ship.

There is a need to read labels of this type more easily.

OBJECT

It is an object of the present invention to provide an improved identification label of machine-readable form, and/or an improved set of instructions for reading labels of this type, or to at least provide the public with a useful choice.

STATEMENT OF THE INVENTION

The various aspects of the invention can be ascertained from the claims. For example in one aspect the invention provides a machine-readable label as defined in claim 1, in another aspect the invention provides a set of labels as defined in claim 4 or in another aspect the invention provides a method of identifying one or more items at a site as defined in claim 9.

In another aspect the invention comprises a computer-readable label or symbol for identification purposes, comprising a matrix of computer-readable indicia; each indicium containing at least one element of readable information, and in which the matrix of information includes an error-correcting code and at least one indicium serves as a locator to provide a reading machine or computer with the location of the remainder of the matrix.

Preferably the error-correction code is a cyclic code capable of coping with individual bit errors.

More preferably the error-correction code is the BCH code, or a derivative of it, such as a shortened BCH code.

Preferably the label is optically readable and accordingly each of the computer-readable indicia is provided with one of an optically detectable, defined range of specific levels of brightness. 30217 PC17NZ96/00021

Preferably the label is capable of scattering or reflecting light, and accordingly brightness levels are equivalent to reflectance levels.

Accordingly the label is, for the purpose of reading, capable of being illuminated by a transducer-compatible source of electromagnetic radiation.

Preferably the resolution or density of pixels in the image-collecting device of the reading machine is such that an area of at least three by three of the sampling elements used by the reading machine is provided to at least partially cover each indicium.

Preferably the label is surrounded on at least one side by an outer edge composed of a relatively bright surface which is preferably at least three sampling units wide.

Preferably this surface corresponds to the white level of the grey scale.

Preferably at least two sets of indicia serve as locators.

Preferably the locators are, on detection, capable of providing a reading machine with information defining the location (that is, position and/or orientation) of the remainder of die matrix.

Preferably each such locating indicium comprises at least one matrix element surrounded by a field of contrasting matrix elements.

Optionally the matrix element or elements of any one locating indicium may also be used to indicate a step or steps of a scale of brightnesses.

Optionally no locating indicia are used; in which case the reading machine attempts to correctly decode the matrix of indicia by a process of repeated trials at different locations and orientations.

Preferably each cell of the matrix of computer-readable indicia is optically detectable and preferably each indicium has a determined reflectance.

Preferably each cell of the matrix of computer-readable indicia is provided with one of two contrasting reflectances.

Optionally each cell of the matrix of computer-readable indicia is provided with one of more than two contrasting reflectances.

Optionally, therefore, any cell may be composed of a material having an intermediate level of reflectance.

Preferably the or each intermediate level is evenly spaced between the brightest and darkest levels.

Alternatively the optical characteristics of a range of cells providing different levels in a scale may include a type of reflectance which is perceived as a contrasting level by an array of sensors comprising more than one set, each set having a response pattern dependent in a different way on the characteristics of the reflected energy.

Optionally a full range of colours may be used.

Optionally each cell may comprise an array of a plurality of dots of one contrasting surface placed upon a substrate of another contrasting surface in variable proportions in order to simulate intermediate levels of reflectance.

Preferably each cell is large enough that when the anticipated environmental "noise" is superimposed on it, the preferred error correction process sees the interference as random noise rather than burst noise affecting adjacent pixels.

Preferably the matrix of information-carrying indicia is read in a consistent order so that in use each indicium comprises a predetermined part of an array of information, incorporating means for detecting and correcting any errors.

Preferably the error-correcting code is a shortened BCH code.

Optionally the error-correcting code is a full BCH code.

Optionally any other error-correcting code capable of replacing the intended information in the event of corruption of cell brightness may be used.

Preferably a printing device is provided with a generator of printable patterns according to the ternary shortened BCH code information protocol of this invention so that a series of unique labels, compatible with the computer reading process, can be generated.

A preferred generator of printable patterns comprises a computer capable of receiving a string of characters and converting them into a matrix of cells together with error-correcting code, as described previously in this section.

Optionally the generator of printable patterns may provide a translation of the design and error-correcting codes into a language or form suitable for use in a printer to actually produce the image.

A preferred language is "Postscript".

A preferred label comprises at least a damage-resistant substrate and a display surface capable of holding the indicia of the label.

Preferably the label also includes bar-code and human-readable indicia.

In another broad aspect the invention includes a reading machine which is capable from time to time of capturing an image of a field of view, which may include one or more labels, and holding the image internally in a form comprising a matrix of sampled points each of which may be mapped to a corresponding point within the image.

Preferably an illumination device is provided to flood a field of view with electromagnetic radiation at least during the period of capturing the image.

Optionally a scanning illumination device may be used to illuminate a field of view, in a sequential fashion.

Preferably the reading machine includes a solid-state camera electrically coupled to at least one addressable memory plane accessible to a digital computer operating under a stored program, and includes an output interface. /30217 PC17NZ96/00021

Preferably the output from the reading machine comprises the information contained within the or each label.

Optionally the reading machine may be physically separated into an image collection portion and an image analysis portion, separated by a communications link.

Optionally the reading machine may be non-optical; that is, it may use microwave radiation or sound (preferably ultrasonic sound) to illuminate a field and collect radiation from discrete sites of the field preferably using phased and/or time-controlled illumination, together with a suitable detector.

Preferably, then, the symbol and its indicia shall preferably exhibit varying yet controlled degrees of reflectance to the non-optical illumination.

Preferably the reading machine locates the position and orientation of any one label in an image containing one or more labels by first detecting the characteristic appearance of a set of one or more accompanying locators.

Preferably it then reads the located data cell matrix.

Preferably the reading machine determines the actual reflectance of the label by a process of determining the most reflective and the least reflective portions and scaling the apparent reflectance accordingly.

Preferably the reading machine compensates for uneven regional illumination on any one label by examining the apparent brightness of the outer edge and compensating the apparent brightness of the adjacent matrix accordingly.

Preferably the reading machine shall be capable of determining the relative levels of the steps of a grey or colour scale, if any, using the range of reflectances included within locating indicia, so that in use any indicium can be assigned to a corresponding and known level in the scale of reflectances in use.

Optionally, the reading machine may be capable of locating the matrix of data cells without the aid of the locators, and in that case it would carry out a series of trial readings until it detected that a reading was valid.

DRAWINGS The following is a description of a preferred form of the invention, given by way of example only, with reference to the accompanying diagrams.

Fig 1: is an illustration of a computer-readable symbol of the present invention.

Fig 2: is an illustration of a combined human-readable symbol array, a bar-code array, and a computer-readable symbol upon a printed label of the present invention, ready for use.

Fig 3 is a block diagram of the processes of the present invention.

Fig 4 is a grey-scale rendition showing the software analysis process for a symbol according to the present invention.

PREFERRED EMBODIMENT This invention comprises a computer-readable label or symbol for identification purposes. This generally optical label has been optimised in particular for applications where (although at the same time the label shall be as small as possible):-

(a) image capture is carried out in possibly adverse circumstances, such as outdoors or where part of the label may be obscured, lost, contaminated, lost in shade, curled up, or the label is presented obliquely or unsharply,

(b) the exact position and orientation of the label is not defined,

(c) there may be a plurality or large number of labels in any one captured image, and

(d) a high degree of accuracy in reading the label is required.

In one example application these labels are used to identify tree trunks, or logs. Each pre-printed label (which simply serves as a unique identifier) is attached to the cut end of a log and an accompanying bar-code is scanned or the alphanumeric characters are noted down. Other data such as quality and ownership are also noted and stored in a master database along with the corresponding label number, so that as the log passes through a series of stations it may be identified and (for example) its ownership may be established. Typically these stations are along a transport chain such as trucking, then shipping, then trucking to a destination sawmill.

At any one of these stations there may be a requirement to read the labels on a bundle of logs in an efficient, quick, safe manner and report their presence at a certain time and place to a management system which includes the master database. Therefore we have developed a camera system and image analysis procedure capable of snapping a picture of the ends of a bundle of logs while held on a truck or in a cradle, detecting all labels, decoding the data, and passing it to the management system. On a wharf, logs are usually placed in a cradle by a lifting machine. A sling is passed around the cradled logs, and they are lifted into the ship. While the logs are in the cradle, they can be photographed.

One main requirement is an error rate (specifically a bad label report error, in which a wrong number is unknowingly delivered), of no more than 1 in 100,000 labels. The most harm is done to the database by incorrect information, rather than by providing a blank. Another conflicting requirement is to use a low-cost and hence low-resolution digital camera, possibly with an imperfect CCD chip, to capture images. A further conflicting requirement is tolerance to damaged or obscured or dirty labels. And another requirement is tolerance to variable levels of lighting.

Therefore there is a need to design a label in which the best use is made of the display area of the label; given that human-readable indicia, and a bar-code, are also required. There is a need to make it plain to the reading computer or machine when an error exists, and to provide correction means if one does exist.

We have provided a 29-bit number, with 17 further bits available for error correction, by using (in the example label) a 6 x 7 matrix of squares, each of which may be coloured white, grey, or black. Each cell is a minimum element of the information carried by the label. This label uses a ternary number base. Some of the 56 cells are reserved, in example labels, for locators (see later) and some are reserved for the site of a conventional barcode. Fig 2 shows an example label as it would be used, in which the "ternary BCH" code 100 of the present invention, together with corresponding alphanumeric descriptors (202, 207), an owner's name and/or logo 206, and a corresponding barcode 203 are provided. BCH stands for a preferred error correction procedure. 205 indicates optional descriptors. 204 is an optional tear line for separating part of the label for separate processing. This label has a border 201 , merely representing the edge of the paper or like substrate for the symbols.

Fig 1 shows the actual data cell section itself. The actual label 100 has no actual border or boundary, though preferably a space at least three sampling units wide which is preferably in the maximum reflective state (i.e. white) is provided about the entire bar code for reconstructing detailed illumination variation, if required. Each unit cell (101,105, 108, 109) is shown here as a dark grey, a speckled, or a white square. (We cannot display squares as black in this specification owing to restrictions on patent drawings - no solid blacks are allowed). Nevertheless, in printed labels the darkest portions are actually black (although as will be elaborated later, a ternary system having three grey levels need not be stretched to either limit of reflectance).

In Fig 1 , the locators are shown as cells 101, 102, 103 and 104 together with the surrounding contrasting space such as the space 107. This space is reserved. The data matrix 106 may extend about the locators. The un-used space between the locators 103 and 104 is reserved in our example label for a bar code 203.

In our trials, camera noise appears to be about + or - 5 units in a 256-step level, and combining this scatter (which may represent noise or internal compensation) with adverse illumination effects in the outside environment leads us to conclude that more than a ternary encoding system would be too likely to result in errors. Of course, in a more controlled imaging situation a greater than ternary encoding system may be adequate. We prefer to err on the side of caution. We see base-4 or base-5 or more cells as being too easily prone to higher errors in our example application, at least, even though the amount of information that can be stored rises markedly as the number of levels goes up. Future versions of this labels may even cover 256 steps in each of red and green and blue channels giving 2^24"1 combinations per cell, if printing and reading technology (to say nothing of fading inks in daylight) allow such resolution in a cost-effective package. A machine version of human-type colour with at least two channels, and perhaps extending to infra-red or ultra-violet) in any one cell, may be used. For the present embodiment, a ternary scale made by black "ink" is preferred.

The label is sized with respect to the image capture device and intermediate factors (distances, focusing elements and the like) so that each cell at least partially covers at least three pixels and preferably has a spatial resolution corresponding to four pixels per cell along both a horizontal and a vertical line, (or sampling elements, applicable if a scanner and A-D converter is used instead) which are internal to the reading machine.

This size requirement allows for rejection of pixels that overlap a transition in the label or possibly their incorporation in more sophisticated analysis, or in label reconstruction should a label image be compromised. (It is not generally possible to repeat a photograph in the target environment). It also allows for some latitude in focusing the label onto the image plane of the sensor, or movement smearing or the like. It also allows some latitude for use of cameras that have defects in their CCD arrays. Such defects are well known and include isolated cell defects, row defects, and/or column defects. (Defect-free cameras exist, but dust spots or crystal imperfections can arise from time to time during manufacture and so defect-free cameras command premium prices).

The external space (see later) is particularly used to determine gradations of illumination over any one label. It should preferably be at least one matrix square (or three pixels of the reading machine's camera) wide, although conveniently it can be wider.

LOCATORS

We prefer to use least two, and preferably four sets of indicia (such as 101 with 107 to serve as locators. Each set or locating indicium comprises at least one cell or matrix element 101 surrounded by a field of contrasting matrix elements 107; and generally the surrounding field is white. We make one pair of adjacent locators 103, 104 assume the black level, and the other two assume the grey level. The software is able to determine the intended reflectance of any cell once it can measure the actual brightness of the locators and their surrounding fields, and thereby compensate for variations in lighting or exposure that may otherwise detract from accuracy.

Preferably the locators are, on detection, capable of providing a reading machine with information defining the location (that is, position and/or orientation) of the remainder of the matrix. The software uses sub-pixel accuracy to determine locator position. Optionally no locating indicia are used; in that case the reading machine attempts to make sense of a postulated matrix of cells by a process of repeated trials at different locations and orientations.

BCH CODE

The preferred BCH (Bose-Chaudhuri-Hocquenghem) code family, developed in 1959 and 1960, can be regarded as a generalised form of Hamming codes for correcting multiple errors. They are cyclic, constructive codes suitable for communication channels in which errors affect successive symbols independently. The well-known Reed-Solomon codes appear to be a special case of BCH codes.

We have developed a shortened (47,29) length BCH code to cope with our preferred rectangular array of data cells (in 100) containing 30 ternary (3-level) cell blocks, which represent 47 bits of binary data. This BCH code has a guaranteed minimum distance of 7, so that at least 7 bits would have to be in error before a code would be miscategorised. The shortened code does not cover all 26 million possible variations in number, yet it is adequate to meet the desired level of accuracy. We use 5 (binary) bits for the alphabetic character and 24 bits for the number, fitted into the 47 bit BCH address space - in a spaced-apart format.

Of course, in some circumstances an error-correcting code may not be necessary. In our application it enhances the reliability of reading damaged labels, yet allows the overall label size to be smaller than if other methods, such as data duplication, were used.

We also lay our matrix out so that straight lines are minimised; we use the so-called Knight's Walk strategy to lay out the cells.

A COMPLETE SYSTEM

Fig 3 shows a data processing system 300 for identifying logs. A device 301 to generate preferably unique patterns of "ternary BCH" symbols including data, an error code, and locators, with other label material (see Fig 2) according to this invention sends data to a printer 302. A preferred printer language is "Postscript". The printer may be located at a forest site, or may simply prepare a large stack of labels for field use. 303 represents a stack of labelled logs, in front of a camera 304 which passes a digital signal to a Wide Area Network interface 305, which transmits the signal to a receiver 306, then to an image analysing computer 307. This reproduces the data 308 originally contained in the ternary BCH symbols.

PRINTING

Preferably, there shall be a device 301 to generate data cell matrices based on the preferred ternary BCH code placed at the site where labels are printed, so that a series of unique labels can be generated. A computer-driven printing machine 302 is preferred, and one kind is a conventional laser printer applying fused toner to a preferably damage-resistant paper, while another preferred kind is a "Printronics" L5024" type that uses an xenon flashlamp to fuse toner onto a substrate which can include vinyl, a material that has a melting point lower than the temperature used in normal laser printers. This machine has the additional benefit that its blacks are matt. Shiny blacks appear white to a camera with a side-mounted flash lamp. Optionally each cell as printed may comprise an array of a plurality of dots of one contrasting surface placed upon a substrate of another contrasting surface in variable proportions in order to simulate intermediate levels of reflectance; a grey level. We use dots that are about 1 mm square, as shown in Fig 1 at 101, for example. As the labels will be illuminated substantially perpendicularly along the optical axis, the blacks should preferably be matt and not reflective.

THE CAMERA

We require a portable data-gathering station that can rove about a transport site such as a wharf; an individual wearing a backpack and holding the camera is envisaged. We prefer to use a battery-driven hand-held solid-state camera 304 containing a CCD chip with an X-Y array of for example 1000 pixels high by 1500 across. A "KODAK" DCS 420 camera (IR version) with 1536 x 1024 pixels is in use. A higher resolution camera would of course have advantages, but cameras of the preferred type are available as off-the-shelf items from several suppliers, and minimal or (by the use of active correction) minimised-defect examples may be found. This camera is assisted by a conventional flashgun, though we prefer to use a flash of infra-red light so that (among other reasons) other people in the yard are not distracted or blinded by the use of the flash, especially at night. Some people are driving 50-tonne log-carrying trucks near the cradles being photographed.

We prefer to use a Wratten 25 filter on the camera, preferably coupled with a filter to block infra-red light beyond about 800 nm; thus admitting the 550-900 nm range to the silicon-based CCD device. This red filter allows aiming by the operator. Red or infra-red light may enhance the contrast between the labels and the background of wood which is the usual background. Ideally, a narrow-band source and a narrow-band filter over the camera lens would minimise the contribution of ambient light to the image, and we are experimenting suitable filters. Because the camera is expected to make at least 240 pictures in a four-hour spell between battery changes it is not practical to use a very bright flash (with heavy batteries) to overcome ambient light.

The camera passes digital image data out for processing as soon as one image has been collected, although as supplied it includes a hard-disk storage device to store about 60 - 70 images. A preferred camera lens is an 18 mm fisheye lens, as this allows the operator to approach the cradle or other holder of logs and illuminate it well with the flash lamp, and minimises focus errors especially those caused by an irregular or uneven object plane. The CCD sensitive area comprises a small central part (about 10 x 20 mm) of the 24 x 36 mm image plane of the modified "Nikon" camera and a normal or telephoto lens would necessitate too great a working distance for this application.

A reading device which scans an area with a pencil beam from a laser may be used in some applications, but the total reading time would tend to be too great for a hand-held device. It may be more suitable for a more controlled environment, such as a fixed camera platform reading labels on steadily moving objects such as railway wagons.

Non-optical images may be used in some applications and for example it might be possible to read suitably created labels inside other structures using microwave or ultrasonic holographic techniques. This may be useful for finding labelled sheets of paper filed within a stack, for example in an office.

RELAYING DATA TO THE IMAGE PROCESSING COMPUTER

The camera output carrying the image does not (in the preferred embodiment) go directly to a memory plane for image analysis; there is an intermediate data transmission step. The camera output may be coupled with other information. For example we digitally encode some variables such as backpack battery condition and transmit this to the base station. The ship and the hold of the ship into which logs from that cradle are being loaded may be written on the cradle so that that data is also recorded. As the image processing computer is relatively large and complex it is housed remotely and data is sent to it over a "Novell" wireless wide-area network system 305 operating at around 2.3 GHz with a 200 Kb/second transmission rate. The image data is (at present) an uncompressed TIFF format file and each image is about 1.5 Mbytes in size, so the transmission time is about 8 seconds. The camera is connected to a dedicated portable PC constructed in a small case. The case, with batteries, is placed in a backpack. It runs on 12 volts, and is fitted with a self-booting program that downloads its operating programs over the network, then transmits image data together with some additional information, such as the log count and computer and camera battery condition, to a base station 306. The log count may be independent, perhaps the operator uses a "sheep counter" or perhaps a stick-shaped device with a counter that also squirts a marker like paint on a log as it is pressed against it.

In time we may develop the image processing computer to such an extent that it is included with the camera and this would have the advantage of more quickly reporting an inadequate photograph.

We may also use software image compression before transmission if it is warranted. The "Novell" networking software takes care of error situations.

IMAGE PROCESSING COMPUTER

At present we use a "Digital Equipment Corporation" Alpha "AXP" computer for image analysis; to extract information corresponding to the labels that may be present within a field of view. This computer 307 completes the process within about 5.5 seconds, while making various reports and drawing lines upon a displayed image at the same time. Lists and traces for debugging purposes are available. A production version would not normally offer an operator any opportunity for intervention, and would therefore be faster. We prefer to compile the same program with different switches so that we can develop special needs for a particular application and then make a production version from the same software.

Input to this computer is via a preferred "Novell" network and the output 308 may also be sent through a network or by modem to a management database located at some distant site. One image processing computer can handle images from several cameras in use at the same transport yard, which may be a dock where several cradles of logs are being loaded at one time.

SOFTWARE

Assuming that an image has been loaded as a copy of the camera pixel array (size from about 1024 x 1024 to about 4096 x 4096 bytes; preferably 1024 high by 1536 wide), the software that distills information from that image can best be described as a series of steps.

We assume that any valid labels will be of a certain approximate size, wherein the image of each blob is about 4 pixels by 4 pixels, in order to include an entire cradle of logs in one field of view. Camera operators are told to stand at about a certain distance from the subject. It will be clear that other sizes can be accommodated by either having the program use a scanning approach, perhaps starting with the last suitable "zoom factor" or by programming in a different constant zoom factor. Our preferred camera (having a limited number (1536) of image pixels across) gives an about 3 mm square (as referred to the label) pixel size.

A larger number of camera pixels will provide (a) smaller labels, or (b) more precise reading with more redundancy of data, or (c) more logs on a cradle, but on the other hand 1.5 Mb of data is quicker to transmit and analyse than 16 or even 4 Mb.

The description below assumes that four locators are used on each label. It will be evident that minor changes to the algorithm will accommodate other numbers of locators.

It will also be evident that there is no search for a border as such as a way of locating labels. Some prior art computer-readable labels rely on borders for location, orientation, and sometimes cell spacing information. The only occasion on which the label edge is - lo ¬

used under our present system is when variations in lighting over the label surface is examined. The locators serve to define the position and orientation of the accompanying information. 5

1. The picture (which may optionally be reconstructed at this point if a defect is recognised) is scanned across perhaps every scan line or row to locate dark blobs of about the right size and edge characteristics. The XY co-ordinates of each located blob are saved.

10

In more detail, a suitable dark blob is one in which the brightness declines from a lighter background level to a darker in-blob level, and remains at the darker level for one or two pixels at least, and then rises at about the same rate. A graph of displacement/ brightness would have a trapezoidal form. The computer may

75 "hunt" by comparing adjacent pixels to find the minimum brightness point to get a first approximation to an x,y point.

The program is quite flexible at this stage about the actual value of a light area or a dark blob, so that varying lighting conditions within a single image can be 20 dealt with. For example a short log bearing a label might be recessed from normal sized logs, so its label would be apparently much darker than the others.

2. The actual centre of each blob is located to sub-pixel accuracy by a centroid-finding procedure in which the mean of the displacement x density

25 product is averaged, in both directions. At this stage some further blob validation is done (such as symmetry, for asymmetrical dark blobs are not like the preferred locators, aa contrasting (usually white) locator border, and the like). An accurate centre determination assists in recovering the data from the central pixels of each cell of the information matrix.

30

3. Neighbouring blobs are now located and attempts are made to form "pairs" - blobs at about the right distance from each other that comply with the expected distances between locators.

35 4. "Triplets" of blobs where the outer blobs are at right angles (approximately) to a centre blob are searched for. /30217 PC17NZ96/00021

- 17 -

5. Rectangles are then formed by trying to find a fourth blob at about the right position from both the outer blobs. Note that we are tolerant of actual rectangle accuracy; it would be more correct to call the formed areas quadrilaterals. Lens errors, labels which are not perpendicular to the optical axis, and some damaged labels (for example) will give rise to valid, though non-rectangular sets of data.

6. Unique rectangles are then formed and a list is stored, while apparently overlapping rectangles and other such errors are put into a "too-hard" stack.

7. Taking each rectangle one by one, a data area is set up within it and the data cells are categorised. Particularly relying on the locators' density values and that of the neighbouring white area, a scale of what is black (i.e. falling within a certain range of densities), what is grey (a distinct and higher range), and what is white (another distinct and even higher range) is constructed.

8. The data cells are then recategorised. Here, a partially shadowed label will become evident. These are also put into a "too-hard" stack.

9. The data cells are read as ternary information in a specific order, after orientation has been established by examining the relative darkness levels of the locators, which indicate orientation.

10. The "easy label" data cell data is then passed to a decoder, which extracts the actual ternary information, applies error correction, and reports the information in the usual decimal form. At this stage, error conditions such as a splash of mud or the like are detected and corrected through action of the BCH error check. In an example:

Data sent to decoder: 1011001 00220002002202211011 0220011 1011021. Error check reveals: 2211011 should be 21 1 101 1 and alters that ternary bit.

Perhaps a spot of mud made the cell look darker.

11. The program now reverts to its "too-hard" stack and attempts to recategorise the data cells. Some of the contents of this stack may not be labels at all - just coincidentally similar sets of pixels. Partially shaded labels are dealt with by

"walking" around the edge, which is always a white strip, and collecting spot densities. Once a shadow has been detected, those data cells within it may be compensated.

Splashes of mud over data cells are reported as well as possible and all versions of "too-hard" labels are then passed to the decoder.

12. Within the decoder, the error correction algorithm, (BCH or similar) is applied although if too many bits are lost the information may be unreadable. The decoder takes account of orientation, as indicated by the relative intensity of the locator blobs.

This procedure is relatively speedy because normally none of the time-consuming processes of image enhancement or image reconstruction, involving neighbour operations, are used. Our digital camera input is substantially free of noise. If an application in which the input data was found to include noise was attempted, averaging routines, ranking routines and the like may be applied at an early stage, or once the label positions have been approximately defined, in order to clean up the picture or areas of interest within it.

Fig 4 illustrates, at 400, output from a development option of the program. The label in this example illustration is inverted and was optically distorted by lens aberrations. The size of individual pixels used in the capture of an image is clearly shown as small squares - this illustration is but a small portion of an entire captured image.

In this illustration the software has picked out four locators; 402, 403, 404 and 407. It has ignored other dark masses which failed to satisfy the criteria for a locator, which criteria include a certain range of permitted sizes, and a contrasting zone 405 surrounding each locator. Dark masses which were ignored include the data matrix 401, the bar-code array 410, and other dark material 409 in the image. The software has verified that the locators belong together, after centroid location of their centres, by constructing a rather crude rectangle (lines and diagonals shown as 408). The software has then identified a matrix of 7 x 6 points, and marked them by crosses 406 overlying the expected centres of cells carrying data. On the computer's display screen, these crosses are in red, green, or blue corresponding to the ternary values (0,1 or 2) detected by reading the value of the underlying pixel. Not shown is the process for allocating a point on a grey scale to each cell, (which might be shown as a frequency histogram) or compensation for uneven lighting.

TRIALS

The entire system has not been field-tested at the time of filing this application. A worst-case trial was arranged to check the error rate, which should be no more than 1 in 100,000. A photographer went to a wharf and took photographs of about 105 defaced bar-coded labels which had suffered some form of field damage, from a selection of about 70,000 labelled logs. We made up a transparent grid with a matrix ruled on it to find out how many cells (had they been printed onto the labels) would have been rendered badly decoded. We assumed use of the preferred BCH code, in which 47 binary bits fitted into 30 ternary bits are placed in a cell array on the label.

If an error in computer reading exists, we estimate a cell error arrival rate of 0.09% for the present size of cells. We estimate the overall average probability of misreporting of data cells after correction to be about 1 in 200,000. As 3-bit error correction tends to raise the overall risk of miscorrection. We prefer to reject a label outright if more than 2, or preferably if more than 1 bit requires correction. This study resulted in a final misreporting error rate (where there was an initial machine-reading error) of about 1 in 200,000, and 99.76% of labels would be read correctly.

VARIATIONS

Most variations have been discussed in the text as the occasion arises, while we may also mention the possibility of using non-rectangular cell markings or indicia: such as more prominent locators, which may be circular or elliptical or of any other shape (perhaps a company logo may be one of these). Reflective locators may be used for higher contrast. Labels printed on a retro-reflective surface (that is, one which returns light substantially back towards its source) may be used, and in that case the intensity of the flash may be lessened which at least assists the person carrying its battery pack.

We would like to again state that a ternary base for the cell information is preferred for this particular application, and binary, or higher bases may be used in other applications.

ADVANTAGES

The advantages of the preferred forms of this invention are:

(1) While maintaining a small area, the symbol or label can be detected, and read accurately in most circumstances.

(2) This applies even after deterioration in the field and image capture in an outdoors environment.

Finally, it will be appreciated that various alterations and modifications may be made to the foregoing without departing from the scope of this invention as set forth.

Claims

CLAIMS:

1. A machine-readable label capable of being applied to an item, said label including a plurality of machine-readable indicia, each of the indicia capable of being distinguishable by machine-reading means from a background field or from neighbouring indicia, wherein said machine-readable indicia comprise: a matrix of machine-readable cells making up a data field; each cell containing at least one element of readable information, and in which the matrix of cells includes an error-correcting code and at least one cell serves as a locator to provide a reading machine with the location of the data field.

2. A machine-readable label as claimed in claim 1, wherein the or each said locator cell is surrounded by said background field.

3. A machine-readable label as claimed in claim 2, in which said one or more locator cells contains orientation information on the orientation of the matrix of cells relative to the location of the or each locator cell.

4. A machine-readable label as claimed in claim 2, in which there are at least three locator cells separated from the matrix field by said background field.

5. A set of machine-readable labels, each label capable of being applied to an item, each said label being as claimed in claim 4, wherein the configuration of the at least one locator cell relative to the data field of each label is similar within the set.

6. A set of machine-readable labels as claimed in claim 5, wherein the information on each label differs from the information on every other label in the set.

7. A set of machine-readable labels as claimed in claim 5, wherein each of the machine-readable cells is provided with one of two contrasting reflectances relative to the reflectance of the background field.

8. A set of machine-readable labels as claimed in claim 5, wherein the error-correction code is chosen from the group comprising the BCH code, and a derivative of the BCH code.

9. A method of identifying one or more items at a site, comprising the steps of: labeling each of said one or more items with a machine-readable label chosen from a set of labels as claimed in claim 6 and said data field contains information relevant to its respective item, recording at least one image of at least one labeled item at the site, converting the at least one image into a machine-compatible form, applying machine vision algorithms to detect the presence and position of the or each locator cell, computing the location and orientation of the data field, detecting and decoding information-carrying indicia within the data field to identify each item.