WO2009133393A1

WO2009133393A1 - Locating inclusions in diamond

Info

Publication number: WO2009133393A1
Application number: PCT/GB2009/050429
Authority: WO
Inventors: Michael Peter Gaukroger; James Gordon Charters Smith
Original assignee: De Beers Uk Limited
Priority date: 2008-04-30
Filing date: 2009-04-28
Publication date: 2009-11-05
Also published as: GB0807849D0

Abstract

A method and apparatus for locating features in a translucent or transparent object, and in particular a crystal such as diamond, is described. The method comprises identifying a window in the object and recording a first image of the window at a first orientation relative to an imaging means. Candidate features are identified in the first image. The object is then rotated relative to the imaging means, and a second image of the window, at a second orientation relative to the imaging means, is recorded. A search is automatically made for the candidate features in the second image. The location of those candidate features which are present in the first image and in the second image can then be identified. Alternatively or in addition, candidate features may be identified independently in each image, and a correlation between the features in the two images can then be carried out.

Description

LOCATING INCLUSIONS IN DIAMOND

The present invention relates to a method and apparatus for assessing the internal properties of a potentially transparent object. In particular, although not exclusively, the invention relates to the automatic detection of inclusions in diamond, and locating the inclusions with reference to measurements through a window of a stone, such as a diamond, without identification by an operator.

The invention may be applied to objects such as diamonds already polished into jewellery gemstones or industrial tooling, or having undergone some preliminary working, but is expected to be particularly useful for assessing objects, such as rough diamonds, that might be polished into gemstones depending on their internal properties such as colour and internal clarity.

Background to the Invention

The market value of a polished diamond depends on its colour, cut proportions, internal clarity and weight. For a rough diamond, the value is determined by the expected value of the optimal pieces that can be cut and polished from it, taking into account the costs involved and with a risk premium to allow for uncertainties in the outcome.

The sources of these uncertainties include a lack of knowledge of the detailed shape and size of the diamond, a lack of knowledge of the colour of the diamond material, and a lack of knowledge of its internal clarity. The clarity of a diamond is determined by the size, number and distribution of inclusions within it. The term "inclusions" is generally used in a broad sense, both herein and in the diamond industry, to cover cracks and other macro-defects, as well as inclusions of non-diamond material or other diamond crystals, that are visible under a given magnification, for instance x10. Even when inclusions are not visible to the naked eye they may interfere with the propagation of light and reduce the diamond's brilliance.

Techniques for determining the external shape of a diamond have been established in the past. Such techniques typically involve the production of a series of images or silhouettes of a diamond obtained from many different directions. The images can then be combined to form a three dimensional map of the surface. Examples of such techniques are described in US 4529305, US 5544254, and US 6567156. However, these documents do not provide information on determining the internal clarity of the diamond.

The internal clarity of the material may not be accurately assessed from external appearance using current methods. This is because the facility of seeing into the object is influenced by the refraction and scattering of light caused by the shape and surface texture of the object.

In principle, some of these limitations may be overcome by the technique of refractive index matching, where the object to be inspected is immersed in a cell containing a liquid of a similar refractive index to the material under inspection. For diamond, however, there are no suitable liquids to match its high refractive index (n = 2.42). In addition, this is a complicated and labour intensive process, and would not be practical for use, for example, in an automated sorting machine.

X-ray microtomography can provide information on both the external shape and internal properties of a diamond. The refractive index of diamond at these wavelengths is much closer to 1 , and this facilitates the investigation of internal microstructure. An example of this is described by SkyScan (www.skyscan.be/next/application0601.htm). However, the technique is too slow to be practical in many applications. In addition, certain internal defects, such as cracks or black graphitic inclusions, both of which are commonly found in rough diamonds, have very poor contrast in x-ray microtomography. Expensive high resolution equipment is therefore required to obtain useful results, rendering the technique uneconomic.

WO 02/46725 discloses an alternative method and apparatus for locating inclusions in a diamond. Each inclusion must first be identified by an operator. The diamond is then translated and rotated so that the inclusion is viewed from a number of different directions. Each time the translation and rotation is carried out an operator must identify the inclusion again. As a result the technique is again slow and impractical for automation. It would be desirable to provide a technique capable of identifying inclusions and locating them automatically (i.e. without identification by an operator). It would further be desirable to produce sufficient information regarding the location and orientation of windows of a stone to be able to monitor the location of inclusions without necessarily performing a complete surface scan, and to be able to identify windows of a stone automatically and distinguish them from re-entrants (concavities) in the surface of the stone.

The Invention

In accordance with one aspect of the present invention there is provided a method for locating a feature in a translucent or transparent object, comprising: obtaining preliminary location information about the location of a first candidate feature in the object. identifying a window on the surface of the object; recording an image of the window at a known orientation and position relative to an imaging means; identifying a second candidate feature in the image; distinguishing between internal and external features on the basis of the orientation and position of the window relative to the imaging means; identifying that the second candidate feature and first candidate feature relate to the same internal feature within the object; and combining the first piece of information with the orientation and position of the window relative to the imaging means to obtain an estimate of the location of the internal feature, allowing for the refraction of light at the window surface.

The preliminary location information may be obtained by any means, but in a preferred embodiment is derived from views seen through a window in a preliminary image or images. Most commonly the internal feature will be viewed through the same window in at least one preliminary image that differs by only a small change in viewing angle relative to the object. Thus it may be considered that the defect is tracked from one image to the next. It will be appreciated that it is also possible for the feature to be seen through different windows in the two images, possibly with a larger orientation difference between images. More generally 'preliminary location information' will be understood to be any data available at the time of analysis. It will be appreciated that, if the image data were obtained in a sequence, such as a video sequence, it is not essential to perform the analysis in the order in which the data were obtained, although this may be convenient. It would also be possible to analyse two or more images simultaneously, forming a combined inference as to the position of the defect.

Furthermore, an image may be obtained containing two windows at different orientations, each providing a view of the same internal defect. In this case it will be understood that one of these views and its window data could be considered to be the preliminary location information, or that they could be analysed simultaneously.

In certain circumstances, such as when examining a polished or partly worked diamond (i.e. a rough diamond which has had one or more facets polished on it or which has been sawn), it would be necessary to take other window data into account. This occurs when internal reflections from these windows provide an indirect light path from the defect to the imaging means.

To obtain the most accurate estimate of the position of the defect it is desirable to combine information from many different viewpoints in a manner that provides most weight to the views that provide the most accurate information. Thus it is desirable to estimate the errors in position associated with each observation.

In accordance with another aspect of the present invention there is provided an apparatus for locating features in an object, comprising: an imaging means arranged to record a first image of a window in the object at a first orientation relative to an imaging means; rotating means for rotating the imaging means and object relative to one another so that the imaging means can record a second image of the window at a second orientation relative to the imaging means; processing means arranged to identify candidate features in the first image and second image and that can distinguish between internal and external features using information about the orientation and position of the window relative to the imaging means; wherein the processing means is further arranged to identify the location of those candidate features which are present in the first image and in the second image.

Using a knowledge of the change in orientation of the window, together with its initial location and orientation and the refractive index of the object, the possible locations of the candidate features in the second image can be narrowed down when the locations in the first image are known. This makes it possible to track features from one image to the next. The location of each feature may then be identified, in terms of its offset from the axis of rotation and depth below the surface of the window, by monitoring the apparent distance moved by the feature between the first and second images and compensating for the refractive index of the object. It will be appreciated that references to tracking features between "subsequent" images is not intended to refer to the order in which images were obtained. It should be possible to track a feature from any image to any other image as long as the relative orientations and positions of the window in the two images is known.

It will be appreciated that the technique is applicable to locating cracks and piques inside any transparent or translucent crystal having at least one nominally planar area

(the window) through which the internal features can be observed. The crystal is rotated about an axis, preferably perpendicular to a camera system, and features

(which are preferably much smaller than a typical inclusion) are tracked in the sequence of images. This process may be repeated with the crystal presented to this axis of rotation at different orientations in order to observe the internal features from different angles.

Further preferred features are set out in claim 2 et seq.

Thus defects may be tracked and located by means of their small component parts rather than as a whole or in large parts. This has the advantage that it is particularly applicable to instances in which objects are superimposed on one another. A further advantage is that small features can more readily be tracked across imperfections in the window. The threshold above which features are detected can be a function of standard deviation, making feature detection largely independent of brightness levels in the images.

Preferred Embodiments

Some preferred embodiments of the invention will now be described by way of example only and with reference to the accompanying drawings, in which:

Figure 1A is a schematic illustration of an apparatus suitable for locating windows in a rough diamond;

Figure 1 B is a schematic illustration of an alternative configuration of an apparatus suitable for locating windows in a rough diamond;

Figures 2A and 2B are photographs of a specular reflection of a window and a tangential view of a window;

Figure 3 is a schematic illustration of an apparatus suitable for viewing inclusions through a window;

Figure 4 is a photograph of the window shown in Figure 2, facing the camera;

Figure 5 is the photograph of Figure 4, with candidate features identified;

Figure 6 is a photograph of a diamond showing tracked and untracked features;

Figure 7 illustrates the behaviour of features in successive images;

Figure 8 is a photograph of a diamond with a fragmented window;

Figure 9A is a photograph of a diamond with inclusions identified by clusters of features;

Figure 9B is an x-y-z plot of the features shown in Figure 9A; Figure 1 OA is a photograph of a polished gemstone diamond with an internal pique; and

Figure 1 OB is the photograph of Figure 1 OA including a cluster of features identifying the internal pique.

Prior art location of inclusions in rough diamonds suffer from the problem that the inclusions must initially be identified by an operator. In order to identify inclusions and defects automatically, candidates should be tracked across a number of different views of a stone.

Object tracking is a technique widely used in image processing. A description of this technique can be found, for example, in "Multiple View Geometry in computer vision 2nd edition" by Hartley, R. and Zisserman, A., Cambridge University Press, Cambridge (2003) which provides an account of those aspects particularly relevant to the inference of the nature of a scene from 2D images taken from different viewpoints.

Much of the literature relating to object tracking is of a mathematical nature making use of the technique of projective geometry, a full disclosure of which is contained in Hartley and Zisserman, and which is therefore not reproduced here. This generalised geometry allows a particularly concise description of the geometry of image formation.

Normally, during image tracking, when an object is identified in one image its boundary is tracked using a boundary tracking algorithm, allowing its size, morphology and contrast to be determined. A rectangular mask is generated from the image. The mask is a 2D matrix array, usually having width and height the same as the object to be tracked, and contains elements whose values are based upon pixel levels of a similar rectangular portion of the image in a pattern matching that of the object. The mask is used to locate the object in a second image taken at a different position or presented at a different angle.

This approach does not work well when attempts are made to track inclusions and defects in diamond, because the contrast changes rapidly across different viewpoints. Instead, a mask of a fixed, p re-determined size is used. This is usually much smaller than the size (in the image) of the inclusion to be tracked, and generally only a few pixels larger in linear dimension than the smallest detectable features. Unlike normal object tracking, location of the boundaries of the object takes place after tracking of the feature in the image sequence has been completed.

Window location

Rough diamonds typically have planar, or nearly planar, regions on their outer surface which are sufficiently clear of imperfections that light can pass through them, enabling a viewer to see into the interior of the stone. These can be considered to be

"windows". Such windows result from the natural growth planes of the crystal.

Inclusions in the diamond can be viewed through the windows, but in order to locate the inclusions, the location and orientation of the windows must be accurately determined since light is deviated by refraction at these window surfaces.

Windows may be identified by any convenient technique. Typically this will be carried out by the analysis of a sequence of images of the object from different viewpoints with respect to the object. These may be the same images that are used later to identify features within the windows, but this is not essential.

One way to locate windows is to make use of existing techniques for producing a surface model of the diamond. One such technique is described in GB 2081439. Planar regions in the surface model are likely to correspond to windows in the stone, but this association is not perfect. Concave regions in the surface of the object may be mistakenly identified as windows, and regions of the surface with a stepped texture provide a plurality of fragmented windows which may be either completely overlooked or assigned to one larger window at the wrong orientation.

An alternative preferred technique involves making use of the fact that light will be specularly reflected off a generally planar surface, producing a bright image of that surface, if the source, surface and observation direction are aligned appropriately. This technique, embodied into an optical or reflecting goniometer was first used to measure the angles between crystal faces by Wollaston in 1809. Since the technique forms an image of the crystal surface which may be conveyed onto an electronic sensor for analysis it allows an image of the window regions of a surface of a given orientation to be obtained directly, and will not be confused by concavities or stepped texture.

Any form of optical or reflection goniometry may be used, but it is desirable to apply a method whereby the whole surface of the object, or a substantial portion of it, may be examined for windows in a reasonable time, and with mechanical motions of economic cost and preferably in an automated manner. Without limitation, techniques such as making use of an extended light source or sources rather than a single point source, reducing the number of axes of rotation to be considered and performing a survey scan using a fast detector such as a photomultiplier to identify the bright flashes from the surface may be employed.

Figure 1A shows a simple system for identifying windows by searching for regions of specular reflection. A diamond 1 is mounted on a movable stage including a spindle 2 which enables it to be rotated about axes R1 , R2, R3, and translated in the x, y and z directions. A collimated light source 3 is located on the axis R2. A photodetector 4 measures light 6 emitted from the diamond at 90° to the axis R2. When a window 5 of the diamond 1 is at 45° to the axis R3 of the photodetector, the light from the source 1 will be specularly reflected off the window and into the detector 4, resulting in a peak. It will be appreciated that the stage need not have six degrees of freedom to obtain the specular reflection in the detector 4, but it is potentially useful later in the procedure for six degrees of freedom to be available. The stage may include a spindle 2 enabling rotation about R2, the spindle itself being translatable and/or rotatable about R1 and R3. Alternatively the light source 3 and/or detector 4 may be translatable. The detector 4 may include a camera system, in which case an image of the window 5 can be obtained.

Figure 2A illustrates an image 22 of a specular reflection from a window 25 of a rough diamond 21. The diamond is then rotated through 90° about R2, and a second image 23 recorded by the camera, as shown in Figure 2B. The light source 3 may be moved from the position shown in Figure 1 to a position behind the diamond, i.e. generally on the axis R3, for obtaining the second image 23. The second image 23 shows the tangent to the window 25. This provides the user with sufficient information to deduce the equation of the window plane.

Figure 1 B illustrates an alternative configuration for identifying windows by searching for regions of specular reflection. The diamond 1 is again mounted on a movable stage enabling rotation about axes R1 , R2, R3 and translation in the x, y and z directions. The collimated light source 3 is replaced by a semicircular light source 13, and in one embodiment this may be extended by mirrors (not shown) to provide an almost circular light source extending right around the diamond 1. A photomultiplier 14 is provided opposite the camera 4. The photomultiplier 14 detects reflected intensity peaks as the diamond 1 is rotated about R2. A peak in a plot of light intensity versus spindle position indicates the position of a window. Once a window 5 has been identified, the diamond can be rotated 180° about R2, so that the window is opposite the camera 4 and an image can be recorded.

Once the location and orientation of the window is known, the diamond can be translated and rotated until the window 5 faces the camera 4 (i.e. the window is normal to the z-axis). This is illustrated in Figure 3. The equation of the window plane is adjusted to take account of the translations and rotations applied to the diamond. The user therefore knows the extent of the window in the x-y plane and the distance from the camera in the z-direction. For viewing or tracking internal defects a diffuse light source is preferred, achieved for instance by use of a diffuse light source 7, with a viewing aperture, between the camera 4 and the diamond 1.

Figure 4 illustrates an image 32 of the window 25 shown in Figure 2, when facing the camera, with the diamond illuminated from behind.

It will be noted that the image 22 of the specular reflection (Figure 2A) includes light regions and dark regions. These arise because, in this example, the window is not a polished, perfectly flat facet. Inclusions may be visible through all parts of the window, but can only be located accurately through the flat regions of the window that are light in the image 22 of the specular reflection. Corresponding parts of the image 32 facing the camera should be labelled as "light" and "dark". For subsequent analysis the region of interest corresponds to the "light" regions only. The coordinates of the light regions of image 22 are transformed in order to map them onto the corresponding positions in image 32. Image comparison techniques may be used to refine the alignment of the images.

Feature location and tracking

The region of interest ("light regions") of the image 32 of the window facing the camera is subdivided into small groups of pixels, which in a preferred embodiment are equally sized squares. The pixels within each subdivided area are subjected to statistical analysis to determine whether or not that area contains a feature (i.e. an area of contrast which exceeds a chosen threshold). The statistical analysis may include determining the standard deviation of the lightness of the pixels in the area. If the standard deviation is above a predetermined value, that area is identified as containing a feature.

Where a feature is identified, a 2D mask array is generated for that area. Each element of the mask has a value which is a function of the lightness value of the corresponding pixel from the original image 32. It is not unusual for several hundred such features to be observed in each image. Figure 5 is an illustration of the image 32 of the window 25, with features identified.

Once features have been identified in the image 32 facing the camera 4, the diamond 1 ;21 is rotated a small angle θι about R1 , with the axis of rotation lying in the plane of the window 5;25. This can be seen as a "rocking" rotation. A second image of the window is obtained by the camera 4. Any features inside the diamond (i.e. beneath the window) will appear to the user to move in the "x" direction (since the diamond is rotated about an axis parallel to the y-axis). The distance they appear to move will be a function of their depth inside the diamond and the refractive index of the diamond itself.

Simple ray tracing techniques and trigonometry combined with Snell's law show that, if a diamond is tilted through an angle θ-[, a feature a depth d beneath the window will appear to the viewer to have a position X₁ (the "x-offset") in the x-direction given by:

where x₀ is the distance, in the x-direction, of the feature from the axis R1 about which the diamond is rotated when the window 5 is facing the camera 4, and n is the refractive index of diamond (n = 2.42 for light in the visible spectrum).

It would therefore be expected that any feature identified in the first image 32 which actually corresponds to an inclusion or defect inside the diamond, or a feature at the surface, will move in the x-direction in a fairly predictable manner when the diamond is rocked through an angle

Thus the search for features in the second image can be simplified by using information provided by the distribution of features in the first image. For each feature identified in the first image, a much smaller region of the second image (i.e. small x- displacements only) is examined to see if the feature is still present. A figure of merit (FOM) is generated for each position in which the mask could be located in this smaller region, by performing a matrix operation using the mask array and the corresponding array of pixels from the second image. The position of the mask having the best FOM is taken to be the position of the feature in the second image.

A check is made as to the quality of match of the location of the feature in the second image. This quality check is performed by combining the FOM with another statistical function (e.g. standard deviation again) of the pixel levels in the selected small area of the image. In other words, a similar check is made to that made in the first image when features were identified in the first place. The match is either accepted or rejected, depending on the quality of match.

If the match is accepted, then a new mask is generated (as before) based on the pixel levels within this best fitting area of the second image. The diamond is then rotated again about R1 and a third image obtained, and the new mask is used to locate the feature in the third image in the same way as that described for the second image. This process is repeated for each of the features originally identified in the first image. Each time the diamond is rotated, the features already identified can be searched for in small regions, chosen on the basis of FOM and match quality checks, and new masks generated. This process is repeated for fourth, fifth, sixth images etc. Features can thus be tracked as the diamond is rotated. The fact that the mask for each feature is redefined each time the diamond is rotated makes the process tolerant of changes in feature contrast and shape with rotation.

In the second, third and subsequent images, an additional search is made for "new" features by subdividing the whole region of interest into small areas and using a statistical function to determine whether each area contains a feature, in the same way as previously described for the first image. These new features can then be tracked through later rotations. In a preferred embodiment the diamond is rotated in 5° steps.

Figure 6 is an exemplary photograph through a window 65 of a diamond 61 (different from the diamond 21 of Figures 2, 4 and 5), showing features tracked from the previous image as black squares 66, and features identified for the first time in this image as white squares 67.

For each feature and its mask, an array of 3D co-ordinates within the diamond is generated corresponding to best fitting positions, using the equation for X₁ shown above. Effectively, simultaneous equations can be determined from the x-offset equation and solved for d and x₀. When features are tracked from one image to the next, a set of curves becomes apparent, as shown in Figure 7. Figure 7 is a graph of x-offset against incident angle for nine series of features Seriesi - Seriesθ which could be followed from one image to the next. Portions of these curves can be fitted to obtain values for d and X₀.

In some cases Series2,Series3,Series5, Seriesθ it is clear that the same feature has been tracked throughout the sequence. In these cases the locus of points is a smooth curve. Any two adjacent points on such a curve can be used to determine d and x₀, since the x-offset equation above can be written twice (for X₁₁^₁ and X₂,θ₂) and solved as simultaneous equations, giving: X₂ X₁ cosθ₂ COs Q₁ d = sin θ₂ SJn Q₁

_A/n² - sin² θ₂ Jn² - sin² S₁

Once c/ is known, it can be substituted back into the x-offset equation to determine Xo.

Although only two points are necessary, it is preferred to use a fitting algorithm with at least four points. This can be repeated many times over the same curve, and the results compared. For example, one inclusion gives rise to the points of series 3 shown in Figure 7. The first four points 731 ,732,733,734 can be used to generate an initial calculation for d and X₀. The four points 732,733,734,735 starting one point further on are then used to generate another estimate for d and Xo, and this sequence can be repeated throughout the curve.

In other cases (e.g. Series 4) it is apparent that one feature has been tracked for four points 741-744, and that the next four points 745-748 in the curve belong to a different feature. Thus where a curve has more than one smooth region, separated by discontinuities, the tracking has "hopped" from one feature to another. A curve such as Series 4 with discontinuities can therefore be used to identify the location of more than one inclusion, with each smooth portion of curve being used to identify a different inclusion.

As an alternative, or in addition, the "speed of movement" between successive images can be used to calculate d and x₀. The speed of movement can be considered to be du/dθ where u is the observed position in the x direction.

Where / is the angle of incidence of a light ray on the window as the ray exits the stone, u = xcosθ + d tani cosθ .

For small angular movements tan/ = sin/ so u = x cos θ + d sin / cos θ . But sin / = sin θ/n so u = x cos θ + d sin θ cos θ/n = x cos θ + d sin 2θ/2n and du/dθ = -xs^'mθ + 2d cos2θ/2n = d cos2θ/n - xs^'mθ du d

Thus— = -(i - 2 sin² θ)- x sin θ . dθ n ^{κ ;}

When θ = 0, du/dθ = d/n and d = n du/dθ .

It will be noted that there is a depth dependent term and an x-offset term in the eq uation for speed of movement. The depth dependent term is zero when (i - 2 sin² θ) = 0 , i.e. when θ = 45°. However, the approximation tan / = sin / becomes invalid at large angles, and depth independence is seen to occur around 50°, rather than 45°.

Thus the speed of movement can be used at angles near 50° to determine x₀ where this cannot otherwise be determined (e.g. when a feature cannot be seen at normal incidence). Once x₀ is known the speed of movement can then be used to determine depth.

It should be noted that features within the diamond can only be tracked through windows. As noted above in the section on window location, the region of interest corresponds only to "light" regions in the specular reflection image. Some rough diamonds have very uneven or fragmented windows, as demonstrated by the window 85 of the diamond shown in Figure 8. When such a diamond is rotated, the inclusions will move in and out of view in the window regions. Features are only tracked when they are visible in the light regions, and no attempt is made to track features as they move into the dark regions. In this case there may be "gaps" in the curves shown in Figure 7. However, it is still possible to calculate the locations of features from incomplete curves, as discussed above.

Aggregation of features

Typically, a few thousand features may initially be identified from the initial image of the window facing the camera. Of these, it is typical for the location of two or three hundred to be identified. The features usually decorate the boundaries of defects

(inclusions) and regions of contrast within defects in the crystal. They also decorate regions having contrast in the window surface itself, and in the bottom surface of the crystal.

Figure 9A is a photograph of a diamond 91 with features 92 identified and located using the method described above. Figure 9B shows the absolute location in three dimensions of the features shown in Figure 9A, together with their projections onto the x-y, x-z and y-z planes.

In this example the inclusions in the crystal, and defects on the surface, are generally an order of magnitude larger than the mask used to identify and track features, and this is a typical scenario. Each defect thus comprises a cluster of features. One of these clusters 93 corresponds to an artefact on the rear face of the diamond. Large clusters

94, 95 correspond to inclusions within the crystal. Although it is possible to identify the location of an inclusion from the location of any one of its associated features, cluster analysis and further image analysis (see for instance the chapter entitled Cluster

Analysis, by A. K. Jain, in "Handbook of Pattern Recognition and Image Processing", 1 , edited by Andrew Young, Academic Press, 2007; ISBN-13: 978-0-12-774560-2; ISBN-

10: 0-12-774560-2) can be performed to locate the defect boundaries more accurately and classify defects in terms of their contrast, size and morphology, and to exclude surface features on the back surface of the diamond seen through the window.

Additional applications

The method outlined above was described with reference to a rough diamond, but it will be appreciated that it can be used to identify defects inside any piece of material with a high refractive index, as long as fairly well defined windows can be used. This is typically true of transparent or translucent crystals apart from diamond. Suitable material also includes polished gemstones (including polished diamonds). In this case polished facets provide perfectly flat windows, but such stones can suffer from the disadvantage that facet boundaries from the rear of the stone can interfere with the feature detection. Furthermore, the background contrast, against which defects must be detected, can change dramatically as the stone is tilted. These problems can be addressed, to some extent, by viewing inclusions through a pavilion facet rather than the table or other crown facet, and by mounting the polished stone in a mount with a high refractive index to reduce the total internal reflection Alternatively, or in addition, the total internal reflection can be allowed for in the ray tracing. Figure 1 OA is a photograph showing a polished diamond gemstone 101 containing a black pique 102. Figure 10B is the same photograph after the method described above has been applied. The pique 102 has been identified as a cluster of features 103.

The method described above involves locating the diamond with one window facing the camera, and rotating the diamond about an axis lying in the window plane. If the diamond is rotated about an axis parallel to the window, but offset a distance rfrom the plane of the window, then the x-offset equation becomes

θ

and the depth of the defect can be determined from two adjacent points on the rocking curve by )

or fitted to four or more points as described above. It is only necessary that the distance r between the rotation axis and the window plane is known. This makes it simpler to set up the apparatus correctly since the axis of rocking rotation does not need to be in the plane of the window.

I n instances where the window is slightly convex or concave, the focussing or defocusing of the surface will produce inaccuracies. However, in this case, the "lens- like" behaviour of the surface could be allowed for in the calculations, using Snell's law of refraction or standard ray tracing equations.

It is also possible to apply the system described above to the situation where the window is not initially perpendicular to the camera axis (i.e. the window plane is inclined relative to the camera). As long as the location and orientation of the window is known, similar analysis can be applied if the diamond is rotated about an axis perpendicular to the camera axis (i.e. the diamond is rotated about R1 in Figure 3, even if the window 1 does not face the camera). Features at the surface of an inclined window will appear to move in the x-direction when the diamond is rotated about R1. For a feature below the surface, the locus will be approximately parabolic. Because the orientation of the window is known, it is possible to calculate the locus using standard ray tracing techniques.

In the most general case, the rotation need not even be about an axis perpendicular to the camera axis. As long as the equation of the window plane, rotation angle and camera viewing direction are known, it is possible to calculate the direction cosines of rays inside the stone which are refracted out of the window towards the camera. Any feature viewed within the region of interest may actually lie at a known point on the surface of the window or on a line, having these known direction cosines, extending from this known point into the stone. The position of any internal point on this line may be determined by taking the 3D coordinates of the surface point and moving a fixed (floating) number of direction cosines in the appropriate direction. A reference point is selected, preferably just beyond the stone. The reference point may be selected to have coordinates which are determined by adding a multiple (e.g. 1000) of the value of the direction cosine for each axis to the coordinates of the surface point.

When the stone is rotated about an axis other than the normal to the window, the direction cosines, in real space, of rays inside the stone which are refracted out of the window will change. The surface and reference points previously described will move to new positions. Knowledge of the rotation axis allows these new positions to be calculated.

In the new view, the surface point will be viewed directly. Rays passing from the reference point will travel inside the stone in the new direction before being refracted, at the exit point, towards the camera. The position of this exit point with respect to the new position of the surface point defines the search direction for the feature to be located in the second view. The mask containing the feature contrast is moved from the surface point to the reference point and a figure-of-merit calculated as previously described. The fraction of the search distance at which the best figure of merit is obtained suggests that the feature is positioned at this same fraction of the distance along the line joining the two points. The line joining the two points will have a total distance of 1000 in the case described above where the reference point has been selected to have coordinates corresponding to the surface point coordinates plus 1000 direction cosines.

It will be appreciated that a rough diamond can be examined through any or all windows available. The more measurements that are taken, the greater the accuracy of the location of the inclusions. It is possible to confirm the location of inclusions by triangulation using measurements obtained through different windows. In addition, inclusions may be visible through some windows but not others.

It will also be appreciated that rocking measurements need not be confined to one axis. Figure 3 illustrates an arrangement in which the diamond 5 is rotated about an axis R1 which is perpendicular to the z-direction. Further measurements can be obtained by rotating the diamond about the axis R2, which is also perpendicular to the z-direction. Although the axis R2 is not shown in Figure 3 as lying in the plane of the window 1 , the diamond can be translated to ensure that it is in this plane. An alternative way of achieving similar results would be to rotate the diamond about the camera axis R3 and repeat the measurements with rocking about a "new" R1.

In addition, the illumination can be varied to suit the circumstances of the diamond being examined. A more directional light source is generally required to obtain a specular reflection from a window, but images of inclusions inside a diamond may be easier to obtain if a diffuse light source is used to illuminate the diamond. Such diffuse light could be essentially from a single general direction (in a manner similar to that shown in Figure 3) or the illumination could be approximately uniform in all directions, or anywhere between these extremes. Visible radiation may be used with a camera sensitive to visible radiation. Optionally, different wavelengths could be used for different aspects of the illumination. Alternatively, an infra-red source could be used with a camera sensitive to radiation in the infra-red band.

It will be appreciated that variations from the above described embodiments may still fall within the scope of the invention. For example, the detection of features has been described with reference to identifying features in one image, obtaining a second image, and searching for features in that second image. It will be appreciated that it is possible to obtain a whole set of images and search for features independently in all images before looking for a correlation between features in subsequent images. It will also be appreciated that there is no reason that any of the images analysed needs to be an image of a window perpendicular to the camera axis. Features may be tracked through images in any order.

Claims

CLAIMS:

1. A method for locating a feature in a translucent or transparent object, comprising: obtaining preliminary location information about the location of a first candidate feature in the object; identifying a window on the surface of the object; recording an image of the window at a known orientation and position relative to an imaging means; identifying a second candidate feature in the image; identifying that the second candidate feature and first candidate feature relate to the same internal feature within the object; and combining the preliminary location information with the orientation and position of the window relative to the imaging means to obtain an estimate of the location of the internal feature, allowing for the refraction of light at the window surface.

2. A method as claimed in claim 1 , wherein the preliminary location information is obtained by recording a preliminary image of at least part of the window at a different orientation relative to the imaging means, and identifying the first candidate feature in the preliminary image.

3. A method as claimed in claim 1 , wherein the preliminary location information is obtained by recording a preliminary image of at least part of another window at a known orientation and position relative to the imaging means, and identifying the first candidate feature in the preliminary image.

4. A method as claimed in claim 2, wherein an axis of rotation between the orientation of the window in the image and the orientation of the window in the preliminary image is known, and wherein the location of the internal feature is identified, in terms of an offset from the axis of rotation and depth below the window surface, by monitoring the apparent distance between the second candidate feature in the image and the first candidate feature in the preliminary image and compensating for the refractive index of the object.

5. The method of claim 2, 3 or 4, wherein likely positions for the second candidate feature in the image are estimated from a combination of the location information of the first candidate feature and the orientation and position of the window relative to the imaging means when the image is recorded, and wherein a search is made for the second candidate feature in those likely positions in the image.

6. A method as claimed in claim 5, wherein estimation of likely positions of the second candidate feature in the image includes searching in sub-regions of the image calculated on the basis of the difference between the orientation of the window in the image and the window or other window in the preliminary image.

7. A method as claimed in claim 5 or 6, wherein searching for the first candidate features in the preliminary image includes searching for regions of high contrast.

8. A method as claimed in any of claims 2 to 7, wherein the first candidate feature in the preliminary image is identified by subdividing a region of the image corresponding to at least part of the window or other window into sub-areas, and applying a statistical function to each sub-area to determine whether it corresponds to the first candidate feature.

9. A method as claimed in claim 8, wherein the application of the statistical function to each sub-area includes determining the standard deviation or variance of the lightness of pixels in that sub-area.

10. A method as claimed in claim 8 or 9, wherein a 2D mask array of elements is generated for the sub-area corresponding to the first candidate feature, each element having a value determined from the pixel value of a corresponding point from the sub- area.

1 1. A method as claimed in claim 10, wherein the search for the second candidate feature includes: determining possible sub-areas of the image which correspond to possible locations for the second candidate feature in view of the difference between the orientations of the window in the image and preliminary image; comparing the elements of the mask with the pixel values of each possible sub- area; and confirming the location at which the mask elements and pixel values of the sub- area match most closely as being the location of the candidate feature in the second image.

12. A method as claimed in any of claims 8 to 1 1 , wherein each sub-area is smaller than the size of a typical inclusion in the object.

13. A method as claimed in any preceding claim, wherein further images of the window are recorded at different orientations of the window relative to the imaging means, and candidate features are tracked between subsequent images.

14. A method as claimed in claim 13, wherein the apparent position of candidate features is tracked across multiple images, and the real location of each feature is determined from the apparent position in three or more subsequent images.

15. A method as claimed in any of claims 2 to 14, wherein the difference between the orientation of the window relative to the imaging means in the image and the preliminary image is defined as rotation of the object about a first axis parallel to the plane of the window.

16. A method as claimed in claim 15, wherein the first axis is in the plane of the window.

17. A method as claimed in claim 15 or 16, wherein searching for second candidate features in the image includes searching along lines perpendicular to the first axis.

18. A method as claimed in claim 15, 16 or 17, wherein a further set of images of the window is recorded, and changes in orientation of the window relative to the imaging means between subsequent images correspond to rotation of the object about a second axis parallel to, or in, the plane of the window.

19. A method as claimed in any preceding claim, further comprising distinguishing between internal and surface features on the basis of the orientation and position of the window relative to the imaging means.

20. A method as claimed in any preceding claim, wherein inclusions are identified as clusters of features.

21. A method as claimed in any preceding claim, wherein further images are obtained of one or more other windows in the object, and locations of candidate features are identified from the further images, wherein the locations of candidate features obtained through different windows are correlated.

22. A method as claimed in any preceding claim, wherein, when the window is fragmented into panes, candidate features are only tracked between subsequent images when they are visible through the panes of the window.

23. A method as claimed in any preceding claim, wherein the location and orientation of the window is identified by positioning the window so that a specular reflection therefrom is directed into the imaging means.

24. A method as claimed in claim 23, wherein the object is rotated 90° relative to the imaging means so that a tangential view of the window is obtained.

25. A method as claimed in claim 23 or 24, wherein the object is illuminated by a light source with an axis at 90° to an axis of the imaging means, so that a specular reflection of the window is recorded by the imaging means when the window is at 45° to the axis of the imaging means.

26. A method as claimed in claim 23 or 24, wherein the object is illuminated by a light source extending at least partly around the object.

27. A method as claimed in any preceding claim, wherein the object is a crystal.

28. A method as claimed in claim 27, wherein the crystal is a diamond.

29. A method as claimed in claim 28, wherein the diamond is a rough diamond.

30. A method as claimed in claim 28, wherein the diamond is a polished diamond.

31. A method as claimed in claim 28, wherein the diamond is a partly worked diamond.

32. An apparatus for locating features in an object, comprising: an imaging means arranged to record a first image of a window in the object at a first orientation relative to an imaging means; rotating means for rotating the imaging means and object relative to one another so that the imaging means can record a second image of the window at a second orientation relative to the imaging means; processing means arranged to identify candidate features in the first image and second image and that can distinguish between internal and external features wherein the processing means is further arranged to identify the location of those candidate features which are present in the first image and in the second image.

33. An apparatus as claimed in claim 32, wherein the processing means is arranged to identify the location of each feature by monitoring the apparent distance moved by the feature between the first and second images and compensating for the refractive index of diamond to calculate the depth beneath the window.

34. An apparatus as claimed in claim 32 or 33, wherein the processing means is arranged to search for candidate features in the second image by searching in sub- regions of the second image calculated on the basis of the difference between the first and second orientations.

35. An apparatus as claimed in claim 34, wherein a candidate feature is identified by identifying a region having contrast which exceeds a chosen threshold.

36. An apparatus as claimed in any of claims 32 to 35, wherein the processing means is arranged to subdivide a region of the image corresponding to the window into sub-areas and apply a statistical function to determine whether each sub-area corresponds to a candidate feature.

37. An apparatus as claimed in claim 36, wherein the processing means is arranged so that application of the statistical function to each sub-area includes determining the standard deviation or variance of the lightness of pixels in that sub- area.

38. An apparatus as claimed in claim 36 or 37, wherein the processing means is arranged to generate a 2D mask array of elements for each sub-area corresponding to a candidate feature, each element having a value determined from the pixel value of a corresponding point from the sub-area.

39. An apparatus as claimed in claim 38, wherein the processing means is arranged so that the search for each candidate feature in the second image includes: determining possible sub-areas of the second image which correspond to possible locations for the candidate feature in view of the difference between the first and the second orientations; comparing the elements of the mask with the pixel values of each possible sub- area; and confirming the location at which the mask elements and pixel values of the sub- area match most closely as being the location of the candidate feature in the second image.

40. An apparatus as claimed in claim 39, wherein the processing means is arranged to retain each candidate feature in the second image only if the pixels in the sub-area of the second image satisfy a statistical function.

41. An apparatus as claimed in any of claims 32 to 40, wherein the processing means is arranged to conduct an additional search for candidate features in the second image without reference to the candidate features identified in the first image.

42. An apparatus as claimed in any of claims 32 to 41 , wherein the imaging means is arranged to record further images of the window at different orientations of the window relative to the imaging means, and wherein the processing means is arranged to track candidate features between subsequent images.

43. An apparatus as claimed in claim 42, wherein the processing means is arranged to track the apparent position of candidate features across multiple images, and the real location of each feature is determined from the apparent position in three or more subsequent images.

44. An apparatus as claimed in any of claims 32 to 43, wherein the rotating means is arranged to rotate the object about a first axis parallel to the plane of the window.

45. An apparatus as claimed in claim 44, wherein the first axis is in the plane of the window.

46. An apparatus as claimed in claim 44 or 45, wherein the processing means is arranged so that searching for candidate features in the second and subsequent images includes searching along lines perpendicular to the first axis.

47. An apparatus as claimed in any of claims 44 to 46, arranged so that the imaging means records a further set of images of the window, in which changes in orientation of the window relative to the imaging means between subsequent images correspond to rotation of the diamond about a second axis parallel to, or in, the plane of the window.

48. An apparatus as claimed in any of claims 32 to 47, wherein the processing means is arranged to identify inclusions as clusters of features.

49. An apparatus as claimed in any of claims 32 to 48, arranged to: obtain further images of one or more other windows in the object; identify locations of features from the further images; and correlate the locations of features obtained through different windows.

50. An apparatus as claimed in any of claims 32 to 49, arranged so that, when the window is fragmented into panes, features are only tracked between subsequent images when they are visible through panes of the window.

51. An apparatus as claimed in any of claims 32 to 50 and comprising an illumination means for illuminating the object, and arranged so that the location and orientation of the window are identifiable by positioning the window so that a specular reflection therefrom is directed into the imaging means.

52. An apparatus as claimed in claim 51 , and arranged to rotate the diamond 90° relative to the imaging means so as to obtain a tangential view of the window.

53. An apparatus as claimed in claim 51 or 52, wherein the illumination means is a light source with an axis at 90° to an axis of the imaging means, so that a specular reflection of the window is recorded by the imaging means when the window is at 45° to the axis of the imaging means.

54. An apparatus as claimed in claim 51 or 52, wherein the illumination means is a light source extending at least partially around the object.

55. An apparatus as claimed in any of claims 32 to 54, wherein the object is a crystal.

56. An apparatus as claimed in claim 55, wherein the crystal is a diamond.

57. An apparatus as claimed in claim 56, wherein the diamond is a rough diamond.

58. An apparatus as claimed in claim 56, wherein the diamond is a polished diamond.

59. An apparatus as claimed in claim 56, wherein the diamond is a partly worked diamond.

60. An apparatus arranged to carry out the method of any of claims 1 to 31.

61. A method of evaluating a diamond using an apparatus as claimed in claim 60.