Use of Image Processing to Aid Furnace Set-up in Electrothermal Atomic Absorption Spectrometry† PHILIPPE R. BOULOa , JOHN J. SORAGHANa, DARAN A. SADLERb , DAVID LITTLEJOHN*b AND ANDREW CREEKEc aSignal Processing Division, Department of Electronics and Electrical Engineering, University of Strathclyde, Glasgow, G1 1XL UK bDepartment of Pure and Applied Chemistry, University of Strathclyde, Glasgow, G1 1XL UK cUnicam Atomic Absorption, York Street, Cambridge, CB2 2PX UK A charge-coupled device (CCD) camera has been used to (GFTV) fitted to a SOLAAR 939 atomic absorption spectrometer( Unicam Atomic Absorption, Cambridge, UK) .The obtain high-definition video images of the cross-section of the tube in a graphite furnace atomizer.Image processing imaging system, which is shown schematically in Fig. 1, consists of a computer-controlled movable mirror, a fixed plane mirror, algorithms have been developed to extract automatically from the images certain features which are of use during the initial a focusing lens and a CCD camera.The system is situated between the furnace head and the monochromator and is not set-up and alignment of the atomizer in an atomic absorption spectrometer. These features include the alignment of the tube, visible to the user. To view the cross-section of the tube, the movable mirror is rotated to divert the undispersed light from the angle of a platform with respect to the vertical axis and the height of an autosampler capillary tip above the tube wall the hollow cathode lamp (HCL) onto the CCD via the plane mirror and the lens.The HCL provides illumination for the or the platform. In addition, algorithms have been developed to analyse images recorded after sample injection, to detect system and any object placed inside the tube, such as a platform, the capillary tip or the sample volume, will cast a whether all of the liquid has been deposited correctly into the tube. Detection of the poor injection of a blood serum sample shadow onto the CCD.Hence, the presence of an object in the tube is associated with an absence of signal on the image is given to illustrate the usefulness of the algorithm. recorded by the CCD camera. Keywords: Electrothermal atomic absorption spectrometry; The camera is a Pulniz type TM 580 (Pulniz Europe, charge-coupled device camera; image processing ; atomizer Alzenau, Germany), which consists of an interline transfer alignment CCD with 500×582 pixels.Each pixel is 12.7×8.3 mm, giving an optically sensitive area for the CCD of 6.35×4.8 mm. The It is well known that the accuracy and precision of an analysis output from the camera is a standard composite video signal. by ETAAS are significantly affected by the initial alignment A Video Blaster SE frame grabber (Creative Labs, Milpitas, and set-up of the instrument. For example, incorrect alignment CA, USA), situated in a personal computer, receives the video of the furnace head increases signal noise because a greater and displays the images on the computer monitor.The software intensity of emission from the tube wall passes into the to display the GFTV images, which runs under Microsoft spectrometer. Also, incorrect positioning of the autosampler Windows 3.1 (Microsoft, Seattle, WA, USA) is incorporated, capillary tip in the tube causes poor repeatability of injection, and may be run concurrently with the SOLAAR software.which adversely affects the precision of the AA signals. To Examples of images obtained from the GFTV system are perform tasks such as adjustment of the capillary tip height shown in Fig. 2. In Fig. 2(a), a platform and the autosampler and monitoring of liquid injection, a dental mirror is often capillary tip can be seen, whilst Fig. 2(b) shows a poorly used to view the interior of the tube. This is not an ideal aligned platform. procedure, as the image of the tube cross-section is small and Images obtained with the GFTV were saved to floppy positioning of the mirror is often inconvenient. Observation of diskette in an 8-bits-per-pixel (256 grey levels) tagged image the tube is simplified if a charge-coupled device (CCD) camera file format (TIFF). These images were then loaded into the is used to obtain images of the tube cross-section.In this VISILOG (Noesis Orsay, France) image processing package, paper, a number of image processing algorithms are described in which all of the algorithms were developed and tested.which can automatically determine from the tube images a Additional code was written in the interpreted C language number of features which are useful for instrument alignment which accompanies the VISILOG package. The time required during the set-up of the atomizer. Furthermore, by comparing images obtained prior to injection and after injection it is possible to determine the profile of the sample droplet.Hence, it is possible to detect the presence of residual liquid on the autosampler capillary tip, the movement of liquid around the tube wall under capillary action, or any contact between the sample droplet and the tube when using a platform for atomization. If uncorrected, all of these problems can cause poor accuracy and/or precision in analysis by ETAAS. INSTRUMENTATION The images shown in this paper were all obtained with the CCD camera system known as graphite furnace television † Presented at the Eighth Biennial National Atomic Spectroscopy Fig. 1 Schematic diagram of the GFTV set-up. Symposium (BNASS), Norwich, UK, July 17–19, 1996. Journal of Analytical Atomic Spectrometry, March 1997, Vol. 12 (293–300) 293Fig. 2 Typical CCD images showing (a) a capillary tip and platform and (b) a misaligned platform. to process an image using the VISILOG package on a personal (1) The image is converted from an 8-bit grey-level image computer with a Pentium processor is approximately 2 s.into a binary image, containing only zeros and ones. All the pixel intensity values below a threshold value are set to zero and all other pixels are set equal to one. The value of the IMAGE ANALYSIS PROCEDURE threshold will vary depending on the HCL and current used The complete image analysis system consists of two distinct for illumination and hence must be determined at the beginning sections. The first section analyses images during the instru- of each set-up procedure.The equation used to determine the mental set-up, the second section provides information on the threshold is given by: deposition of a liquid into the tube and the presence of any residual sample left on the capillary tip. Each section consists threshold= p�+pmax+pmin 3 (1) of an appropriate pre-processing stage, followed by an analysis stage. The resulting information is then presented to the user. where p� is the mean value of the pixel intensities, calculated Instrumental Set-up over the whole image; pmax is the maximum pixel intensity and pmin is the minimum pixel intensity.Fig. 4(a) and (b) shows an Overview example of a GFTV image before and after thresholding, The instrumental set-up analysis flow diagram is shown in respectively. Fig. 3. It was designed to identify the following key features: (2) Isolated points around the edge of an object, which may (1) Alignment of the furnace and camera.(2) Type of tube cause problems in detecting the boundary edge, are removed. (straight-edged or part-ridged) and presence of any platform. This is achieved using a morphological filter, such as an (3) Alignment of the tube (if part-ridged) or platform to the openingoperator.1 This is a smoothing operator which discards horizontal axis. (4) Presence of the capillary tip. (5) small objects in the image, but keeps the largest ones with a Characteristics of the capillary tip, i.e., depth and position. shape similar to the original object.Fig. 4(c) shows the effect The first function is to determine the presence of a platform. of the opening operator on the GFTV image shown in This is required as different algorithms are used to assess the Fig. 4(b). The individual, detached pixels around the edge of alignment of the furnace and camera, depending on the pres- the image of the tube wall, visible in Fig. 4(b), are removed by ence of atform in the tube. Once the alignment of the the opening operator and are not visible in Fig. 4(c). Further furnace has been checked, the angle to the horizontal of any analysis of the CCD images is then carried out on the platform can be determined. If no platform is present then the pre-processed image. type of tube (straight-edged or part-ridged) is identified. This information is used by the instrument to ensure that the optimum heating rate, which is tube-type dependent, is applied. If the tube is part-ridged it is possible to use the ridges to determine the angle of the tube to the horizontal axis, which Detection of platform is a measure of the correct alignment of the tube in the To find the boundary of an object in a given direction, the atomizer.If the tube is not accurately aligned in the furnace image data are scanned in this direction and the boundary head, the user can be warned and the tube re-aligned. Once edges of an object are given by a transition from zero to one the type and alignment of the tube have been determined, the or from one to zero.The detection of the presence of a platform location of the autosampler capillary tip is found. The first is based on scanning the image vertically. The first transition function is to determine whether or not the tip has actually from one to zero, encountered when scanning from the top of entered the tube. Once the capillary tip has entered the tube, the image, at three equidistant positions across the tube is the depth of the tip and the angle, to the vertical axis, is found denoted by A, B and C.This is shown schematically in Fig. 5. and displayed numerically to the user. Hence, it is possible to A platform is present only if these three points lie on a straight know the exact depth of the capillary for any particular line. The choice of the horizontal separation of the three points, analysis and to reproduce accurately the same depth at a Dx in Fig. 5, is obviously important. If Dx is too small, the later date. points A, B and C may appear to lie on a straight line, even if no platform is present. However, if Dx is too large, points Pre-processing outside the platform may be detected and hence A, B and C may not lie on a straight line, even if a platform is present. A Prior to image analysis, all images undergo a pre-processing satisfactory value for Dx has been found to be 0.4 mm for a step, the primary aim of which is to obtain an image which is easier to analyse.It consists of two separate steps: tube diameter of 5 mm. 294 Journal of Analytical Atomic Spectrometry, March 1997, Vol. 12Fig. 3 Flow chart of the overall instrument set-up system. Fig. 4 (a) Original CCD image of a tube; (b) binary image obtained after thresholding the image in (a); and (c) image (b) filtered by an opening operator. Journal of Analytical Atomic Spectrometry, March 1997, Vol. 12 295Fig. 5 Procedure for detection of the presence of a platform.(a) A, B and C are not aligned: there is no platform; and (b) A, B and C are aligned: there is a platform. Alignment of the f urnace head When the furnace head and the camera are not aligned properly, the image of the tube cross-section is non-circular. Thus, the extraction of quantitative information from the CCD images may not be accurate due to the inability to relate the physical dimensions of the tube to the image dimensions. The Fig. 7 Schematic diagram showing the two methods used to detect user can be made aware of this situation and decide whether the alignment of the atomizer with the CCD camera. (a) Method 1: case of cuvette with platform; and (b) Method 2: case of straight or or not to re-align the furnace head. Fig. 6 gives real GFTV ridged cuvette. Left: the camera and furnace head are aligned; and images showing images from aligned and misaligned furnace right: the camera and furnace head are not aligned.heads, respectively. It is clear from Fig. 6(a) that, in the aligned case, the CCD image of the tube cross-section is a circle whereas in Fig. 6(b), the misaligned case, it is an ellipse. aligned and misaligned atomizers the cross-section of the tube is assumed to be circular if 86°<a<94°. Detecting the alignment of the furnace head is then equivalent to detecting whether the image is circular or elliptical. Method 2. This method has been designed to detect the Depending on the presence of a platform, a different method alignment of the furnace and the camera when the tube is was devised.The two different methods are illustrated in Fig. 7. straight or part-ridged, and no platform is present. It is illustrated in Fig. 7(b). Eight points along the boundary of the Method 1. This method, shown in Fig. 7(a), is used when a platform has been detected. The method uses Apollonius’ top of the object are found by scanning the image horizontally at four different levels.The circle, C1, fitting these eight points theorem,2 which states that if parallel chords of an ellipse are drawn along a direction other than along the axes of the is computed. This operation is repeated on the bottom of the object to obtain the circle C2. For a normal, straight-edged ellipse, and their centres joined by a line d, the line d is perpendicular to the direction of the chords only if the ellipse tube, the image of the cross-section is circular if all 16 points lie on the same circle.Hence, if the length l between the centres is circular. The angle a between the line d and the chords is therefore computed to check the alignment of the tube. If of C1 and C2 is computed, then l#0 implies a correctly aligned furnace head. This method also works with part-ridged tubes a#90°, the furnace head is therefore aligned. This method is satisfactory only when the chords span the area of the ellipse because the centre of the circle defining the inner diameter of the tube and the circle defining the edge of the ridges are as much as possible; therefore, it is necessary to use the information already obtained about the platform position to concentric, even though the diameters are different.Even for a perfectly aligned atomizer the value of l may not determine the lowest position on the tube edge which may be used to construct a chord. To compute the angle a, five parallel be exactly equal to zero due to noise and imperfections in the imaging process. A lower limit is therefore required which, if chords are constructed at equidistant points along the direction given by d, and linear regression of the chord centres is used exceeded, indicates that the tube cross-section is significantly non-circular.From the examination of a number of images, a to determine a. From the analysis of a number of images of Fig. 6 (a) Example of correctly aligned head and (b) incorrectly aligned furnace head. 296 Journal of Analytical Atomic Spectrometry, March 1997, Vol. 12lower level for l of 10 pixels (0.3 mm) was considered appro- platform to the horizontal is then the tangent of the slope of this line. Using this procedure an analyst can assess conditions priate. In Fig. 6(a), for example, the value of l is 8 pixels, which is below the threshold and therefore indicates a well aligned to give an adequate sample injection, which may typically be a platform angle of, for example, no more than 2° to the atomizer.However, for the misaligned case, shown in Fig. 6(b), the value of l is 39 pixels, indicating that the tube cross-section horizontal axis. For Fig. 2(a), the angle of the platform to the horizontal axis was calculated to be -0.6°, which is satis- is elliptical. factory, whereas the platform in Fig. 2(b), which is clearly misaligned, was calculated to be at an angle of -6.5° to the Determination of tube type horizontal.The method for determining the type of tube is illustrated in Fig. 8. The two circles C1 and C2, computed during the furnace head alignment process, are used in this step. The radii of the Detection and characteristics of capillary tip two circles are compared, and if the two circles have the same To detect the presence of the autosampler capillary tip, the radius, the tube is a normal, straight-edged tube, whereas if CCD image is scanned horizontally at a number of different the radii are different, the tube is part-ridged.vertical positions, and the number of boundary points for each When the tube is part-ridged, the alignment of the tube is scan, n, is counted. If n is equal to 2, only the edges of the computed by finding the intersections of the ridge with the tube have been detected and there is no tip at this level. inner diameter of the tube. The image is scanned along a circle However, if n>2, one of the boundary points corresponds to whose centre is the centre of C2 and whose radius is the a boundary of the tip, implying the presence of the tip at this average of the radii of C1 and C2.The intersection points, i.e., position. The end of the tip can be found by using a binary points A and B in Fig. 8, are detected as boundary points recursive bisection method,3 which is a computationally fast along this circle. The angle of the tube to the horizontal is the search procedure which eliminates the need to scan the angle of the line AB joining the end points of the ridges.whole image. When the position of the capillary tip end is known, the Determination of the angle of the platform depth and the angle of the tip to the vertical axis may be computed as shown in Fig. 10. The depth of the tip is easily The angle of the platform can be approximated by calculating calculated by scanning the image vertically from the end of the angle of the line passing through the points A, B and C, the tip to the bottom of the tube, or the platform when it shown in Fig. 5. To obtain a more accurate measure of the exists, and by calculating the distance between these two angle, more points along the top edge of the platform are points. By knowing the internal diameter of the tube, the depth found by scanning across the platform. This is done by of the capillary can be accurately calculated. With a tube producing an orthogonal projection from the centre of the diameter of 5 mm the position of the tip end can be calculated tube cross-section to the platform, using the approximate value to approximately 0.1 mm.For Fig. 2(a), the capillary tip was of the angle of the platform as a guide. The image of the calculated to be at a height of 0.8 mm above the platform. platform is then scanned either side of this projection to The angle of the tip is obtained after having found two determine the boundary points. This is illustrated in Fig. 9, points on each side of the tip.Scanning the image horizontally where five equidistant points are shown scanned across the just above the end of the tip gives two points on each side of centre of the platform. Linear regression of these points is then the tip, labelled A and A¾ on Fig. 10, and the two others, used to give the best line fitting these points. The angle of the labelled B and B¾, are found by analysing the image along a circle whose centre is the centre of the tube and whose radius is smaller than the radius of the tube.The angle of the tip to the vertical axis is then the mean of the angle of each line, i.e., AB and A¾B¾, joining the detected points on each side of the tip. The tip is considered central when the position of the end of the tip is on the central vertical axis of the tube, i.e., d=0 in Fig. 10. Fig. 8 Detection of the tube type and the tube angle. Fig. 10 Characteristics of the autosampler capillary tip. Fig. 9 Computation of the angle of the platform.Journal of Analytical Atomic Spectrometry, March 1997, Vol. 12 297Sample Injection Analysis System General algorithm The flow diagram for the algorithm is shown in Fig. 11. Two The second component of the CCD image processing software analyses images taken after injection to determine whether the images are required, the first is a reference image which contains an image of the capillary tip prior to injection. The liquid has been deposited correctly into the tube, and if there is any residue left on the capillary tip.The GFTV system can second image is of the tip at the same position, but after injection of the sample. Each image is then pre-processed and record images during the sample injection and drying phase of the furnace programme and then the movable mirror is the two images are subtracted to produce a difference image. This image is then further processed, resulting in an image of automatically rotated out of the optical axis.Hence, images of each injection in an analysis sequence can be obtained. the sample left in the tube. Pre-processing the CCD images The two CCD images are first pre-processed using the same method previously described, i.e., using a threshold operator and an opening operator. In order to remove objects in the image of the capillary tip, due to transmission of light, further processing is required. In this discussion an ‘object’ is taken to mean a collection of adjacent pixels whose value is equal to one.Firstly, an erosion operator1 is used to remove any small objects. The erosion operator works by moving a disc, of diameter d, around the inside of the boundary of an object. If the disc is entirely included within the object, the boundary point is replaced by the centre of the disc. The effect of such an operator is to remove small objects (of a size <d), to shrink other objects, to discard peaks on the object boundary and to disconnect some parts of the object.The parameter d is chosen to be equal to half of the width of the capillary tip. Small regions of transmitted light within the image of the capillary tip will then be removed in the eroded image, however, the size and shape of the image of the capillary tip, the sample droplet, the tube wall and any platform will be distorted. This image is then used in a reconstruction operation1 as a ‘marker Fig. 11 Flow chart of general analysis system for sample injection.image’. A reconstruction operator requires two images, the Fig. 12 Processing of a capillary tip image to remove the effects of transmitted light. (a) Raw unprocessed CCD image; (b) tip image after processing by thresholding and opening operator; (c) tip image after erosion; and (d) reconstruction of image (b) using image (c) as a marker. 298 Journal of Analytical Atomic Spectrometry, March 1997, Vol. 12original image and an image containing the markers.The to induce capillary action and cause the sample to flow up the outside of the tip. These raw images are then pre-processed reconstructed image is formed from the original image by including only those objects whose boundary contains the and subjected to the erosion and reconstruction operators, the results of which are shown in Fig. 13(c) and (d). The logical position of a marker in the second image. The marker may be a single pixel or a collection of pixels. This operation keeps subtraction of the two images, Fig. 13(e), results in an image which shows the presence of the serum sample around the end objects in the original image which are larger than d, so that the size and shape of the image of the capillary tip, the sample of the capillary tip and also around the injection hole.The use of the opening operator cleans up the subtracted image by droplet, tube wall and any platform remain unaffected. The effect of the erosion and reconstruction operators on a reference removing the individual pixels around the inner diameter of the tube, and it is this image, Fig. 13(f ), which is shown to the tip image is shown in Fig. 12, where it can be seen that the transmitted light in the original reference tip image has been user. In Fig. 14, an example of a good injection is shown. The raw images of both the reference image and of the capillary removed, without degrading the overall shape of the image of the capillary tip. Both the reference image and the image after tip after injection, have been omitted.The capillary tip was adjusted to the correct depth and the delivery of the serum sample injection are subjected to this pre-processing. sample into the tube was acceptable. The subtracted image, Fig. 14(c), is completely blank except for some pixels around Sample visualization process the inner diameter of the tube. Once the opening operator has been used the resulting image, Fig. 14(d), is completely blank.To determine the position of the liquid in the tube, the logical The sample cannot be seen as a part-ridged tube was used and subtraction on a pixel-by-pixel basis of the pre-processed image the sample is lying below the height of the ridge. (after injection of the sample) and the pre-processed reference image is computed. The resulting difference image is processed, with an opening operator, to remove pixels due to any small CONCLUSION difference of position between the two images.The final image is an image of the position of the liquid in the tube and any A number of image processing algorithms have been developed for use with a CCD camera system that provides cross- residual liquid left on the tip. This process is illustrated in Fig. 13. A reference image, of the tube with the capillary tip sectional images of a graphite furnace atomizer tube. The algorithms are designed to provide the user with information prior to injection, and an image after injection of a blood serum sample with the capillary tip at the same position as in which is useful when initially setting up the atomizer and spectrometer prior to an analysis.The alignment of the furnace the reference image, are shown in Fig. 13(a) and (b), respectively. The capillary tip was deliberately set too low in the tube head to the CCD camera is checked, which is important to Fig. 13 Processing of images for sample visualization, poor injection of blood serum.(a) Raw capillary tip reference image; (b) raw capillary tip image after injection; (c) pre-processed version of image (a); (d) pre-processed version of image (b); (e) logical subtraction of images (c) and (d); and (f ) image (e) after application of an opening operator. Journal of Analytical Atomic Spectrometry, March 1997, Vol. 12 299Fig. 14 Processing of images for sample visualization, good sample injection. (a) Pre-processed capillary tip reference image; (b) pre-processed capillary tip image after injection; (c) logical subtraction of images (a) and (b); and (d) image (c) after application of an opening operator.minimize the passage of continuum emission from the tube used. Both of these effects will lead to problems during an analysis. The algorithms developed here can be used to warn wall into the monochromator. Furthermore, a measure of the the user when a poor injection has occurred, or to instruct the angle to the horizontal axis of a platform is provided, which instrument to disregard the signal and repeat the measurement. can ensure that the platform is correctly aligned. For optimum control of the furnace heating, the type of tube used is automatically determined. A quantitative measure of the depth REFERENCES of the capillary tip in the tube is also obtained. This is a useful 1 Serra, J., Image Analysis and Mathematical Morphology, Academic measure as the analyst can set the depth to an accuracy of Press, New York, 1983. #0.1 mm, which means that once the optimum depth for a 2 Bronshtein, I. N., and Semendyayev, K. A., A Guide-book to particular sample type has been found, the capillary tip can Mathematics, Verlag Harri Deutsch, Frankfurt, 1971. be reset to exactly this depth each time the sample type 3 Press, W. H., Teukolsky, S. A., Vetterling, W. T., and Flannery, is analysed. B. P., Numerical Recipes, Cambridge University Press, In addition, algorithms have been developed to analyse Cambridge, 1992. images recorded during injection with an autosampler. These may be used to monitor the liquid delivery and to provide Paper 6/06255E warnings if, for example, there is residual liquid on the tip, or Received September 10, 1996 Accepted December 10, 1996 sample in contact with the tube wall if a platform is being 300 Journal of Analytical Atomic Spectrometry, March 1997, Vol. 12