locating defect 1

Upload: renukadevi1

Post on 07-Apr-2018

222 views

Category:

Documents


0 download

TRANSCRIPT

  • 8/6/2019 Locating Defect 1

    1/25

    LOCATING DEFECT ON SHIRT COLLARUSING IMAGE PROCESSING

  • 8/6/2019 Locating Defect 1

    2/25

    Introduction:Recent developments in the hardware and software makes

    the automation of visual fabric inspection tasks is becoming feasible atlow cost. There are problems in repetitive, tedious and physically demanding humainspection for defects in shirt collars .

    The above problem can be solved by using Image Processing(the automatof visual fabric inspection).

    Image processing techniques gave an indication of fast computationtime to detect the defects on the image , which is needed in manufacturing,and could be applied to most automated inspection systems.

  • 8/6/2019 Locating Defect 1

    3/25

    Description

    The primary purpose of inspection of shirt collars is toavoid collars with faults, which might affect the saleability of the product.

    One of the advantages of using a modern ChargeCoupled Device (CCD) image sensing device is coverage of alarge area.

    Faults in shirt collars may be due to faults in the originalcloth, or might arise during manufacturing.

  • 8/6/2019 Locating Defect 1

    4/25

    SHIRT COLLARAfter the shirt collar pieces are cut ,each fabric panel is

    divided lengthwise into two areas, the collar band region andcollar region , along the folding line.

  • 8/6/2019 Locating Defect 1

    5/25

    First panel collar regions (seen during wearing of thegarment): Collar Point Left (CPL). Collar Point Right (CPR). Collar Back Centre (CBC).

    Second panel collar band regions (most exposed inpresentation windowing): Right Collar Band (RCB).

    Left Collar Band (LCB). Central Collar Band (CCB).

  • 8/6/2019 Locating Defect 1

    6/25

    USE OF HIGH QUALITY VISUAL

    APPEARANCE:A high quality visual appearance is of importance in two stages:

    (1) the presentation of the product in themerchandising pack (known as windowing );and(2) during the consumer use of the product.

  • 8/6/2019 Locating Defect 1

    7/25

    The fault types are:

    Th ick weave : an extra piece of loose yarntrapped in the weave.S lub flaw : a short thick place in the yarn

    where the fibres are not spun properly.Knot : appears like a prominent knot or spoton the surface of the fabric.

    M is-weave : caused by incorrect interlacingbetween weavers and wrappers, which leavesloose threads over the surface of the cloth.

  • 8/6/2019 Locating Defect 1

    8/25

    Ladder : faults due to threads missing from the weftor warp.H oles : gaps in the cloth larger than the usual spacingbetween the yarns.Colour flaw : a short thick piece of a yarn of dissimilarcolour trapped in the wrapper or weaver.Foreign fibre : a foreign fibre, i.e. a strand or piece of fibre of dissimilar colour (usually black), dragged intothe yarn during the spinning process.

    B lack mark : parts of the yarn, either wrapper orweaver.S tains : marks caused by heavy contamination of fabric by oil or grease or other substances.

  • 8/6/2019 Locating Defect 1

    9/25

    METHODOLOGY

    HIGHLIGHTINGMETHOD

    STATISTICALTECHNIQUES

    CONTROL ANDSENSITIVITY

    VARIANCEMETHOD

    MOVING

    GROUPAVERAGE

    SIGNATURECOUNTING

  • 8/6/2019 Locating Defect 1

    10/25

    Image capturing environment

    The CCD camera and back-light were used in this research. The captured image was digitised to 256 x 256pixels with 256 grey level (zero representing black and 255 maximum brightness or white).The camera focus was set at 5mm and aperture set at 5.6 for an optimum depth of field and suitableexposure. The area of the image was chosen to be 2.3in. by 3.7in. (58.42mm ( 93.98mm ) The field of viewcaptured could be varied by altering the height of the camera. The height of the camera (lens to objectdistance) was 10in., giving an effective pixel area of (height x width) = 0.11mm x 0.18mm.

    S PECIFICATION S:

  • 8/6/2019 Locating Defect 1

    11/25

  • 8/6/2019 Locating Defect 1

    12/25

    Inspection system

    In order to detect shirt collar defects, thesystem acquires static images of a movingobject on a carrier using CCD cameras.The PC system controls the image processingand finds the defect. When a fault is detected,the host computer interrupts the motor driver.The defective shirt collar is then removedfrom the carrier. This could be done with arobot or any pick and place device.

  • 8/6/2019 Locating Defect 1

    13/25

    Detection methods

    Statistical techniquesA statistical method for automatic fabric inspection, attempts to

    reduce the amount of computation and data handling associated withstatistical techniques by using grey level signatures for horizontal andvertical rows. Horizontal and vertical signatures work because mostdefects tend to be elongated longitudinally and laterally.

    The signature for a vertical column or horizontal row of an image issimply the sum of the grey levels in the pixels of the column or the row .

    The mean and standard deviations of these signatures are found inthe usual way

    N is the sample size of the column or row in pixels, which in thisapplication is 256.

    The presence of a fault causes a significant deviation in thesignatures of the columns or rows in which the fault occurs.

  • 8/6/2019 Locating Defect 1

    14/25

    Control and sensitivityThe designer must consider the sensitivity; k is the constant of

    sensitivity. This is usually different in plain fabric than wool fabric. Trial anderror is the only way to judge the sensitivity. The constant of sensitivity onall the shirt collar application was same, then operator use the softwarewith the same control limit.

    The following control limits are defined: Vertical upper control limit Vertical lower control limit

    Horizontal upper control limit Horizontal lower control limit

    Using mean and standard deviation of the column or row signatureled to the development and design of two novel approaches: movinggroup average, and moving divided group average.

  • 8/6/2019 Locating Defect 1

    15/25

    M oving group average :A program was designed using mean and standard deviation of the column or rowsignature to allow the user to capture or load the image, and to select different k

    values and different group numbers of columns or rows.Calculating the moving average of groups, in each image there are 256 columns(rows). Each signature is the sum of 256 pixels.Grouping every n, n = 1, 2, ..., 256, columns (rows) and taking their average, gives256/n groups. Each group can be moved by any number of steps m, m = 1, 2, ...,256, then the next group average calculated, and so on.

    The result of this grouping and averaging will produce a graph. If the graph liesabove or below or touching the two levels, an alarm will sound to void thedefective fabric.

    If a graph not exceeds the upper or lower levels means there is no defect and noalarm will sound. If a graph exceeds the upper or lower levels which means thereis a defect and an alarm will give notice of it.

  • 8/6/2019 Locating Defect 1

    16/25

    M oving divided group average :The column or row is divided into a number of groups as a power of 2, then themoving average method explained previously is applied.

    The graph here is more precise than the graph in moving group average, because

    the graph here is elongated to 512 points since every column is divided into twogroups.If the result of the collar image exceeds the upper or lower level then there is adefect and an alarm will give notice of it. This graph gives a more precise resultthan moving group average.

  • 8/6/2019 Locating Defect 1

    17/25

    Highlighting method :The object is separated from the background by choosing a suitable

    threshold (T) value where T = [mu] or T = [mu] + [sigma] or the mode (themaximum or the minimum frequency).

    The best automated threshold value for this application can be calculatedfrom the grey level.

    Applying this method separated only 60 per cent of the defects from thebackground, so the variance filtering was used .

  • 8/6/2019 Locating Defect 1

    18/25

    Variance method :This method is done by sliding a 3 x 3 window along the whole image. Any

    other window size can be used, but a 3 x 3 window is appropriate for thisapplication because it gives the best result.

    If variance value >= R (reference deviation) the pixel is set to white;otherwise, the pixel is set to black.

    Applying the variance method over the complete image separated 90 percent of the defects from the background.

    Then the moving group average and signature counting method wereapplied .

  • 8/6/2019 Locating Defect 1

    19/25

    Detection methods with filtering

    The variance method was used on the defect collar image, to separate thedefect from the background.Then, the moving group average method was used to show the shape of the defect.

    The vertical and horizontal signatures were counted( signature counting )to represent the defect number.

  • 8/6/2019 Locating Defect 1

    20/25

    The moving group averageThe moving group average is applied on the signature S, mean [mu] andstandard deviation [sigma]. There are three graphs.

    The upper graph represents [mu] + [sigma], the lower graph represents[mu] - [sigma], and the middle graph represents the mean under theimage. The three graphs,clearly shows the defect and its shape.

  • 8/6/2019 Locating Defect 1

    21/25

    Signature countingThe signature of each column or row of the image is calculated, then thelines in the signature graph counted and compared with the reference(non-defect) to give an indication of the defect.If the number of columns in the signature graph of the sample image, andthe number of columns in the signature graph of the reference image(non-defect). Their similarity signature ratio value, must be zero. Themismatch must lie between 0 and 1, where 0 indicates a perfect matchand any number between 0 and 1 a complete mismatch.When we run the variance filter algorithm on a non-defect ca, no defectwill be counted. But Noise and discontinuity effects on the image willaffect the result of the signature counting algorithm. It may be concludedthat the variance filtering and signature counting method are suitable forthe application of shirt collar defect identification.

  • 8/6/2019 Locating Defect 1

    22/25

    Example of optimal filtering : (a) original image of seams withpleated defects and (b) the filtering result.

  • 8/6/2019 Locating Defect 1

    23/25

    Result and Display:

    Three classes of sample images of collar defects :(a) seams without manufacturing defects,(b) seams with

    pleated defects,and (c) seams with puckering defects, and their corresponding binary images (d)to(f).

  • 8/6/2019 Locating Defect 1

    24/25

    Merits

    The main reason for inspecting shirt collars instead of the fabric roll is the difficulty in relocating defectsfollowing inspection, due to fabric stretch. Further, machine vision roll inspection machines are veryexpensive.

    The results show that the statistical methods (moving group average and moving divided group average)can be applied to an automated inspection system to detect faults in shirt collars.The moving group average is beneficial because defects rarely extend the full length of a row or column.

    The novel techniques using variance filtering with the moving group average on S, [mu] and [sigma], orwith the signature counting graph showed the fault and its shape very clearly.The operator can see the defect on the graph and an alarm sounds. This makes the system fully automatedand the operator is in the position of an observer.

  • 8/6/2019 Locating Defect 1

    25/25

    THANK YOU