GB2186365A - Inspecting textile items - Google Patents

Abstract

Textile webs or articles e.g. socks are inspected for defects by forming an image e.g.with a video camera 13, capturing data representing at least salient features of the image e.g. in stores 14 and 16, selecting data from the captured data using a mask function corresponding to a given type of fault, and comparing the selected data automatically with reference data. Large and small scale faults in garments can be detected using appropriately sized masks and by enhancing a digitised image by histogram modification techniques, filtering out low frequencies, averaging several video frames, and by image sharpening techniques. <IMAGE>

Description

SPECIFICATION Processing textile items This Invention relates to processing textile items such as garments.

Currently, matching oftextile items, e.g. sizing of socks, together with inspectionforfaults is essentially done bytrained hand and eye. Especially in the case of garments, inspection is an intricate and repetitive job, and prone on that account two human error. It is also expensive because of the intensive skilled labour requirement.

"Inspection" can cover a wide range of operations, and a single product can be inspected for a wide variety offaults. Faults forwhich garments are inspected include dimensional deviations from set standards, non-inclusion or incorrect positioning of trims and components, non-inclusion of stitched seams, incorrect pattern matching, and any ofthe faults that might be detected in the fabric orfabrics thatgoto make upthe garmentwhich could arise in the manufacture, processing, storage and making up.Fabric itself can be inspected for such things as surface distortions from thin areas resulting from thin yarn, dropped stitches, tuck stitches, broken needles, holes, slub cuts, knot cuts, yarn slubs of unacceptable length, incorrect dimensions, presence offoreign material such as broken needle hooks, colour variation, stains, creases, cracks and barre effects. Inspection for fabric faults should idealiy not be left until the garment is inspected, which is wasteful ofthe labour and additional materials content of the made-up garment.

Techniques exist, of course, for detecting some faults otherthan by hand and eye. This places and holes, for instance, can be detected by sensing photo-optically orotherwise light or other radiation transmitted through a fabric. Broken needle hooks can be detected by passing a fabric over a metal detector. There is currently no general method or universal piece of equipment, however, that can be adapted to detect faults of any description.

Butfor many inspection processes there is simply no available automatic equipment of any description. One such process is the sizing of garments especially knitted garments such as socks.

The sock manufacturing process cannot be controlled so exactly that successive items coming offthe same machine are identical in every particular, and this is aggravated where a numberof machines produce items to a common nominal size for later pairing, so that two items selected at random from the common delivery of the machine or machines are unlikely to match closely enough to be regarded as a pair.

The present invention provides methods and apparatus for inspecting textile products which substantially reduce the skilled labour requirements and which are applicable or adaptable to substantially every inspection and selection process at reasonable capital cost and with considerable improvement in reliability.

The invention comprises a method for inspecting textiles comprising forming an imagethereof, capturing data respresenting at least salientfeatures of said image, comparing said data automatically with reference data, said reference data comprising data representative of a fault-free textile, the method being used to detect faults, data being selected from the captured data by a maskfunction corresponding to a given type offault and compared with data selected from said reference data by said mask function, said data being selected by scanning said maskfunction over the captured and reference data.

The image may be formed by an electronic imaging device such as charge coupled device. An image enhancing technique may be used to enhance the quality ofthe image. For example, improved signal-to-noise ratio may be achieved by combining and averaging information from multiple frames of a video image, especiallywhere the object is static at least for long enough to capture multiple frames. A digitised image may be enhanced by histogram modification techniques, image smoothing may be achieved byfiltering out lowfrequency noise, and theimagemaybesharpened by known image sharpening algorithms.

Preferably, an image comprising a 512 x 512 array of pixels is formed.

The entire image is preferably digitised and stored as digital data in RAM locations of a computer which map on to the image. The data can then be processed in a wide variety of ways by appropriate software and data extracted relative to dimensions and faults by appropriate techniques such as edgefinding methods and fault recognition algorithms.

Said maskfunction may, at its simplest, define a 2x 2 pixel fault detection module. Three mask functions may definethree different sizes offault detection module- say 2x2, 4x4 and 8x8 pixels- and these functions may be added together heirarchicallyto produce a variety offault detection modules corresponding to a wide range of length and width of defects.

Astatistic is calculated from each said comparison and tested for its departure from a zero statistic indicative of no difference and hence no fault, a fault being indicated art a certain threshold level of said statistic. Said threshold level may be adjustable whereby to tunethe method to an acceptable compromise between missed faults and false alarms.

Said selection process may comprise selecting items which have acceptable fault levels and rejecting others, and may comprise marking faults in the products.

The invention also comprises apparatus for inspecting textiles comprising imaging means forming an image thereof, data capture means capturing data representing at least salient features of said image, comparison means comparing said data automaticallywith reference data comprising data representative of a fault4ree textile, and selection means selecting data from the captured data and the reference data by a maskfunction corresponding to a given type offault by scanning said maskfunction over the captured and reference data.

Embodiments of methods and apparatus for inspecting textile products according to the invention will now be described with reference to the accompanying drawings, in which: Figure 1 is a diagrammatic plan view of an inspection station for individual garments with a sorting arrangement; Figure2 is a plan view of a sock at an inspection station; Figure3 is a view of a section of video image of an edge; Figure 4 is a diagrammatic plan view of an inspection station for a travelling web offabric, with a Feult marking arrangement; Figure 5 is a representation ofthree masks for use in defect detection; Figure 6 is a representation of how one ofthe masks shown in Figure 4 can be used in a system for detecting different kinds offaults;; Figure 7 is a diagrammatic view of another arrangement at an inspection station.

The drawings illustrate methodsforinspecting textile products 11 comprising presenting such products to imaging apparatus 12 intheform of a television camera to form an image thereof.

The video signals from the camera 11 are digitised in an analog-to-digital converter 13 and selected frames captured in a frame store 14 such as a commercially available Gresham Lion 214 supervisor, which can capture, in real time, images from a standard television camera. The frame store 14 is linked to a minicomputer with 64Kbytes of RAM and 2mbytes of disk storage, and mapping stores 16 and a VDU orTV monitor 17.

Various image enhancing techniques can be used to enhancethe quality ofthe image. Multiple, for example, three frames may be averaged to improve signal-to-noise ratio. The digitised image may be enhanced by histogram modification techniques, smoothed by filtering out lowfrequency noise and sharpened by known image sharpening algorithms.

With such techniques a 51 2x512 pixel array image can readily resolve such small fabric faults as dropped stitches, tuck stitches and presence of foreign material such as broken needle hooks. Atthe same time, larger scale defects such as thin areas, holes, long yarn slubs, incorrect dimensions, colour variation, stains and barre effects can be detected.

Figures 1 and 2 illustrate how dimensional defects can be detected and items sorted according to size.

These figures illustrate the method being used for matching socks 21 supplied as a succession of imperfectly matched socks on atranslucent conveyor belt 22 which presents each sock 21 in turn to the camera 12 which is located over a back lighting panel 23.

It greatly simplifies the measuring algorithms if the socks 21 can be presented to the camera 12 in substantially constant orientation and position. For this purpose a datum line 24 is drawn on the belt 22.

Operatives place the top of the sock on the datum line 24. (The 'top' can be defined either as the upper or lower edge of the welt).

When a sock 21 is in thefield ofview ofthecamera 12, which is of course roughly coincident with the panel 23, the imaging, image enhancing and image storing operations are performed. On the stored image, further algorithms are now brought into operation to detect the edges of the sock. An edge detection method for socks has to satisfy the sizing criteria. If the socks are to be classified in 5mm groups with a precision of + 1 mm on the basis of a full scan size of 500mm, it is necessary to use a statistical estimate of edge position in the image to betterthan a single pixel in a 512x512 pixel image.

One such estimate is Position of edge = P + (E-B)/(O-B) where B = light level background illumination O = light level of object E = light level in edge pixel P = position at beginning of edge pixel.

However, since a sock is not usually a "hard-edged" object, in practice an edge is not seen as a background pixel next to an edge pixel next to an object pixel,- see Figure 3- and it is necessary to sample a number of pixels extending perpendicularly acrossthe edge in ordertocompute the centre of change.

Once the edges have been located, the bisectors of the leg (AB) and foot are (CD) computed, and the position Pwhere they intersect- see Figure 2. The distances AP, PD are useful measures for sock sizing.

If three ranges for each of AP and PD are sorted, each, say, of 5 mm spread, nine different categories are required together with a tenth category for socks unacceptably outside one or both permitted variations. Figure 1 shows a ten gate arrangement with associated blowers 25 underthe control of the computer 15.

Figure 4 illustrates an inspection station like that illustrated in Figure 1, but intended for the inspection of a travelling web of fabric and having a fault marking arrangement 41 instead of the sorting arrangement.

The travelling web 42 passes, again, over a back lighting panel and images formed by the camera 12 captured in the frame store 14. As a fault is detected the computer 15 activates a printer 43 to apply a mark 44to the edge oftheweb 42 where the fault occurs.

The mark 44 can be in the form of an ink-jet printed mark or a seif adhesive label and can indicate the nature of the fault, for example, a thin region, incorrect width, slu b, dropped stitch, or whatever.

Fabric width is measured exactly as previously described in connection with sock dimensions by locating the fabric edges on the image and computing the distance between them. Fabricfaults such as dropped stitches are detected by using a masking technique.

If a 4x2 pixel mask is used, each 4x2 pixel area of the image is examined in turn to detect a departure from the expected image. Such a 4x2 pixel mask is useful for the detection of small defects such as dropped stitches, tuck stitches, presence of foreign matterand soon.

The mask- which is simply a 4x2 pixel representation of what is expected to be seen at any particular location - operates on the representation in the actual (enhanced) image to produce a feature which approaches value zero when there is no defect (i.e. actual isthe same as expected) andthevalue 1 when the mask is coincident with a defect.

As shown in Figure 5, three different masks of different sizes are shown which are useful in the detection of small width, medium width and wide defects.

Asfurtherseen in Figure 6, masks can bearranged in heirarchical tree structure in order to detect larger defects. Thus instead of first searching the image for 2x 2 pixel sized defects, then for 4x 2 defects, then for 8x2, 16x,32x2 defects and so on,the image can be examined oncefor2x2 pixel defects, then the resulting features combined at different levels in the tree structure.

At any node (F1, F2, F3, F(1,2), F(1,2,3,4) etc.) the expected value of the feature is zero for no defect. A null hypothesis of no defect is tested by the deviation of the feature from zero. Thus for a point defect such as a dropped stitch, node F1 might have a feature value approaching 1, while F2, F3, F4 etc. are zero. If, however, F1, F2, F3 the value 1, the feature at node F(1,2,3,4)would approach 1 and this would indicate a line feature- such as a crease- ratherthan a succession of dropped or tuck stitches.

The provision of different mask sizes maximises sensitivity to features of differentwidth. Thus, while the lowwidth maskmighttendto identify a number of adjacent line features as such, this interpretation would be overridden by the wide maskthat might showthesefeatures up as athin area ora hole ora stain depending upon the light level.

The deviation of the feature value from zero is tested by a suitable statistical test such asthet-test for significance. The significance level can be pre-selected to give, in particular circumstances, an acceptable compromise between missed faults and false alarms.

The expected value in a mask is input from a sample offaultfreefabricor afaultfree article used as a standard. Obviously faults are easier to detect in a plain fabric than in a patterned fabric, for which the software would be required firstto "align" the actual and standard sample images.

So far, what have been described are techniques where measurement and fault identification have been effected entirely by algorithmic processes on a captured image. It is possible, however, also to use camera -pointing techniques for measurement, say, in which a servo-controlled scanning camera locates a salientfeature- a toe, for example, of a sock- and records its coordinates, then locates another such salientfeatureand records its coordinates,the dimensions of the item being computed from the coordinates.

Figure7 shows such a system in which made-up garments 71 are passed on a belt beneath a scanning camera 11 having a zoom lens. Each garment 71 is arrested beneath the camera 11 to capture its image.

The computer 15 checks that salient features such as buttons 72 are present and correctly positioned by comparison with a standard reference image.

Having located the buttons,thecomputerthen actuates the servo-motor 73 to point the camera at each button in turn, zooming-in to obtain a high resolution image of the button which is checked against another standard reference image to ensure that the button is properly sewn by ensuring that the thread holesarefilled.

Claims (8)

1. A method for inspecting textiles comprising forming an image thereof, capturing data representing at least salientfeatures of said image, comparing said data automatically with reference data, said reference data comprising data representative of a fau It-free textile, the method being used to detect faults, data being selected from the captured data by a mask function corresponding to a given type of fault and compared with data selected from said reference data by said mask function, said data being selected by scanning said maskfunction over the captured and reference data.

2. A method according to claim 1, in which said maskfunction defines a 2 x 2 pixel fault detection module.

3. A method according to claim 1 orclaim 2, in which three mask functions define three different sizes offault detection module and maskfunctions are added to produce larger and differently shaped fault detection modules.

4. A method according to any one of claims 1 to 3, in which a statistic is calculated from the said comparison and tested for its departure from a zero statistic indicative of no difference and hence no fault, a fault being indicated at a certain threshold level of said statistic.

5. A method according to claim 4, said threshold level being adjustable whereby to tune the method to an acceptable compromise between missed faults and false alarms.

6. A method according to any one of claims 1 to 5, in which said selection process comprises selecting items which have acceptablefault levels and rejecting others.

7. A method according to any one of claims 1 to 6, in which said selection process comprises marking faults on the products.

8. Apparatus for inspecting textiles comprising imaging means forming an image thereof, data capture means capturing data representing at least salient features of said image, comparison means comparing said data automatically with reference data comprising data representative of afault-free textile, and selection means selecting data from the captured data and the reference data by a mask function corresponding to a given type offault by scanning said maskfunction overthe captured and reference data.