Method for determining the color of teeth

Abstract

The invention proposes a method for determining a tooth color, wherein an analysis evaluation device has an iterative learning algorithm using a CNN, which, in a preliminary step, acquires images at different illuminations and shooting angles on the basis of at least one previously known, if necessary virtually generated, sample tooth color and analyzes and evaluates the images and learns a correspondence with the correct sample tooth color. In the analysis evaluation step, a shooting device is provided, with which an image of a support having a previously known, if necessary virtually generated, sample tooth color together with at least one tooth is shot. The shooting device acquires at least two images of the combination of the tooth to be determined and the support from different shooting angles and transfers the images to the analysis evaluation device. The analysis evaluation device analyzes the acquired images on the basis of the learned correspondence with the correct sample tooth color and outputs the tooth color of the tooth to be determined from a reference value, such as a commonly used tooth color chart, for example A1, B2, etc.

Description

TECHNICAL FIELD

[0001] The invention relates to a method for determining a tooth color. BACKGROUND

[0002] It is known from EP 3 613 382 A1 to use a color selector which is held next to a tooth whose tooth color is to be determined. A common image of the tooth and the color selector configured as an auxiliary body is taken. Since the auxiliary body has a known tooth color, the color of the tooth can thus be obtained more easily and more precisely.

[0003] The solution described above requires the possibility of providing reproducible illuminance. Moreover, the detected color of the tooth can be calibrated or normalized due to the presence of the auxiliary body having a known tooth color. It has been shown, however, that despite the possibility, deviations occur in practice, so that the color determination accuracy that can be achieved in the laboratory cannot be maintained in practice. SUMMARY

[0004] In contrast thereto, the object of the invention is to achieve a method for determining a tooth color which ensures that an accurate color determination is also achieved in practice.

[0005] To this end, the invention proposes a method for determining a tooth color, wherein

[0006] - the analysis evaluation device has an iterative learning algorithm using a CNN which, in a preliminary step, acquires images at different illuminances and taking angles and analyzes and evaluates the images on the basis of at least one previously known sample tooth color and learns a correspondence to the correct sample tooth color,

[0007] - in the analysis evaluation step,

[0008] - a taking device is provided with which an image of an auxiliary body having a previously known sample tooth color together with at least one tooth is taken,

[0009] - the taking device acquires at least two images of the combination of the tooth to be determined and the auxiliary body from different taking angles and transmits the images to the analysis evaluation device, and

[0010] - the analysis evaluation device analyzes and evaluates the acquired images on the basis of the learned correspondence to the correct sample tooth color and outputs the tooth color of the tooth to be determined as a reference value,

[0011] - the auxiliary body has a reference point, and in the analysis evaluation step, the analysis evaluation is only started if the reference point is recognized in the image to be analyzed and evaluated, and in the absence of the reference point, another image is requested.

[0012] According to the inventive setting, the analysis evaluation device first collects and analyzes images of the sample teeth at different illuminations and shooting angles in an iterative learning algorithm and learns the correspondence of each collected image to the known correct sample tooth color.

[0013] The collection and learning at different shooting angles according to the invention is important and very helpful for improving the recognition ability of the analysis evaluation device.

[0014] For the image shooting of the sample teeth, a first shooting device is used. This first shooting device can have a particularly high quality. For example, a professional reflex camera can be used in order to be able to read out its raw data. The quality of the raw data is usually better than the data of the camera converted into a standard format, such as JPG.

[0015] But it is also possible to use the camera of an end device, for example the camera of a smartphone, as the first shooting device.

[0016] In the analysis evaluation step, one shooting device is used, for example a further shooting device. In the analysis evaluation step, the image of the tooth whose color is to be determined is shot together with the auxiliary body having a known color.

[0017] The analysis evaluation device uses the learned images and the image data detected with the images.

[0018] The analysis evaluation device preferably has a first part used in the preparatory step and a second part used in the practical use of the analysis evaluation device thereafter.

[0019] The second part uses the same data as the first part.

[0020] It can be achieved thereby that the analysis evaluation device can be used on site, that is to say for example in a dental surgery, without having to forgo the preparatory acquired data.

[0021] In order to achieve easy access to the acquired data and knowledge, it is preferred to store the data and knowledge in the cloud or at least in a region which is protected on the one hand and accessible in the practical use on the other hand.

[0022] At this point, the second part of the analysis evaluation device is used in the practical use, which uses the same data as the first part.

[0023] That is, the data is always available to the analysis evaluation device. But this does not mean that the data is required to be used in full at any time in the method according to the application.

[0024] Rather, it is preferred that the analysis evaluation device has a memory in the second part, the content of which is periodically synchronized with the cloud.

[0025] What the second part of the analysis evaluation device does is to perform an analysis evaluation of the images captured by the - further - photographing device and to correspond the images to the sample tooth colors on the basis of the learned correspondence with the sample tooth colors and then to output the tooth color in the usual tooth shade card on the basis of this correspondence.

[0026] Here the sample tooth colors do not have to be physically present and stored; rather, the sample tooth colors are generated virtually, that is to say it is sufficient to define them in a defined color space. It is also possible to generate a tooth color only numerically in a virtual color space, preferably in the RGB space, and to use this tooth color as a reference. Such a color is referred to here as an RGB tooth color, it being understood here that this also includes colors generated in other virtual color spaces at the same time.

[0027] There are various possibilities for the generation of such virtual tooth colors:

[0028] 1. Use of a scan of a tooth and determination of the corresponding RGB values;

[0029] 2. Use of values from an existing tooth library; or

[0030] 3. Use of a colorimeter, such as a spectrophotometer, in order to determine the color numerically.

[0031] Surprisingly, in the preparatory step, a significantly better recognition and determination of the actual color of the tooth to be determined is achieved by a plurality of photographing conditions.

[0032] The photographing conditions include photographing conditions in which different light sources and brightnesses are worked with.

[0033] For example, the same sample tooth can be photographed with the spectrum of a halogen lamp, an LED lamp and natural light in the sunshine and, on the other hand, in the sky with clouds. Furthermore, this is carried out at three different brightness levels. Furthermore, the photographing is carried out at 5 to 15 photographing angles which differ in the vertical and in the horizontal direction.

[0034] This preparatory work achieves a deep database of, for example, 100 to 300 different photographing conditions for each sample tooth.

[0035] It is understood that the foregoing description is merely exemplary and that the application is in particular not limited to the number of shooting conditions in the preparatory step.

[0036] In an advantageous design variant of the method according to the application it is provided that in the preparatory step, at each iteration, it is checked how much the result changes in each iteration. If the result changes less than a predefined threshold, it can be assumed that the desired accuracy is achieved.

[0037] This design variant can also be improved in that the iteration is terminated only after a number of times below the threshold of the change.

[0038] It is preferred to ensure, after the end of the preparatory step, that all data obtained in the preparatory step, including the correspondence of the results of the preparatory step to the sample tooth colors, are brought to the cloud and, if necessary, the data are accessed in the cloud.

[0039] Before the terminal device performs the analysis and evaluation step, the terminal device performs a data comparison so that the data obtained are stored, in particular completely, on the terminal device, if necessary.

[0040] These data are synchronized periodically so that changes are made available to the terminal device periodically.

[0041] That is, when the terminal device is to perform the analysis and evaluation step, the current data of the analysis and evaluation device are always provided, if these data have been generated and provided in the preparatory step.

[0042] A first part of the analysis and evaluation device works only in the preparatory step, but continues to be available when other illuminations or shooting angles are to be detected.

[0043] In practice, it is reasonable to make fine adjustments, if necessary, although the relevant illumination and shooting angle are detected, in which case the analysis and evaluation device can also be brought to the cloud as an executable or compilable program.

[0044] This solution has the advantage that the dentist himself can use the analysis and evaluation device in the first part of the analysis and evaluation device, if necessary, when the dentist has the necessary equipment, and the dentist himself can take into account his special illumination, provided that the dentist is provided with the sample tooth colors required to perform the first step.

[0045] At this point, the dentist can provide other dentists with new data obtained in this way by the dentist and in terms of the special illumination in the cloud, if desired.

[0046] A similar operation can also be performed when an adaptation adjustment is desired for a region:

[0047] The spectrum of natural light is different geographically between regions near the equator and near the poles, since the absorption bands of the atmosphere play a significantly stronger role in regions near the poles.

[0048] If the average value of natural light is used as a basis in the first step, a dentist in a region near the equator concludes on the basis of the data provided by him via the analysis evaluation device that the natural light data provided has to be corrected.

[0049] The dentist can provide the data adapted by him to the region, for example in the cloud, to other dentists in the region of the dentist.

[0050] This is only an example of a preferred possibility according to the application, namely to realize a data exchange of the provided data in the cloud. It is understood that other data exchanges based on other bases are also included according to the application.

[0051] It is advantageous for the actual analysis evaluation in the analysis evaluation device to provide reference points on the auxiliary body. The reference points are chosen such that they can be recognized by the analysis evaluation device. When, for example, three or four reference points are provided, the arrangement of the reference points in the photographed image can deduce at which angle the photograph was taken.

[0052] At this point, it is advantageous according to the application to draw conclusions from such an angle detection or to compare with data assigned to the relevant photographing angle.

[0053] In another advantageous design, it is provided to divide the photographed image into sections.

[0054] This division has the advantage that regions can be hidden, from which it can be deduced from the data detected in the analysis evaluation step that there is a reflection.

[0055] Surprisingly, the accuracy of the analysis evaluation can be significantly improved by this measure, in particular also in bright and thus highly reflective environments.

[0056] Different parameters can be used for the analysis evaluation by the analysis evaluation device:

[0057] For example, a normalization can be carried out with respect to the ambient brightness, for example by means of an ambient light sensor. Usually, end devices, such as smartphones, have an ambient light sensor, which usually adapts the brightness of the screen.

[0058] In an advantageous design of the application, the ambient light sensor is used in order to carry out the aforementioned normalization.

[0059] High-end smartphones can also distinguish between artificial light and natural light on the light spectrum. This distinction is likewise made by means of a built-in ambient light sensor. According to the application, it is also possible to make use of the result of this distinction, in that inappropriate data is not provided to the analysis evaluation device from the outset. BRIEF DESCRIPTION OF DRAWINGS

[0060] Further advantages, details and features result from the following description of an embodiment of the method according to the application with reference to the drawings.

[0061] wherein:

[0062] Figure 1 a schematic flow chart of the preparatory step according to the application is shown;

[0063] Figure 2 a schematic flow chart of the image pipeline as a subprogram, which is used both in the preparatory step and in the analysis evaluation step, is shown;

[0064] Figure 3 a schematic flow chart of the analysis evaluation step is shown;

[0065] Figure 4 a learning algorithm according to the application is shown as a function with inputs and outputs;

[0066] Figure 5 a schematic diagram of a CNN convolution layer in the algorithm is shown; and

[0067] Figure 6 a schematic diagram of one example of max pooling is shown. DETAILED DESCRIPTION

[0068] The preparatory step 12 of the method according to the application for determining the colour of teeth is shown schematically in Figure 1 The preparatory step 12 is here referred to as training 10 and begins according to the flow chart of Figure 1 the application with the start of the training 10.

[0069] In step 12, images, which can also be referred to as "photos", are first taken. These photos show the sample teeth colours and the respective teeth constituting the sample are taken, captured and analysed under different light intensities and angles of shooting.

[0070] The recorded data are transferred to the image pipeline 14 as a subprogram, which is used both in the preparatory step and in the analysis evaluation step. Figure 2 The design of the image pipeline can be seen.

[0071] After the processing in the image pipeline 14, the data prepared is transferred to the training algorithm 16. The training algorithm carries out the actual training, that is to say the optimal reproduction (Wiedergabe) of the sample tooth color in data form at different illuminations and shooting angles.

[0072] Next, it is checked in step 18 whether the training is sufficient. If this is not the case, that is to say greater accuracy is required, then the return jumps to block 12 and the detected data is passed through the image pipeline 14 again.

[0073] If, on the contrary, the training is considered to be sufficient, then the "trained" data is stored in the cloud in step 20. The training is thereby ended at block 22.

[0074] By Figure 2 The individual steps of the image pipeline 14 can be seen. The image pipeline 14 starts in step 26. There are image data 28, which are stored in an image memory in step 28.

[0075] There is now an image 30 and this is checked in step 32 in terms of its resolution and its format. If the resolution and format are insufficient, then path 34 is entered, whereas if both the format and the resolution are sufficient, execution continues with path 36.

[0076] In path 36, the image 30 is checked in step 38 to find the position of the reference points of the reference target; the reference points are captured. In path 40, the possibility again arises that no reference points or not enough reference points can be determined.

[0077] If, on the contrary, reference points can be found, then execution continues with path 42 and the relevant information is extracted from the image 30 in block 44. According to block 46, the information includes the color values extracted from the reference target.

[0078] It is now checked whether there is a tooth segment in the image 30. If this is the case, then path 48 is entered.

[0079] In parallel to this, the color values are processed with path 50. The color values are transferred to an algorithm in path 54 which generates color information from the color values for the image 30 at block 52, and data transfer is carried out to the calling program in block 56, that is to say the subroutine image pipeline ends.

[0080] According to path 48, the tooth segment data is also processed further. It is checked in block 58 whether there is a reflection. If the reflection is above a threshold value, then path 60 is entered, which ends in the same way as path 34 and path 40 in the case that there is no result according to block 62.

[0081] Conversely, if the reflection from path 64 is below a threshold, the dominant tooth color is calculated using so-called k-means clustering. This is done in box 66.

[0082] The color value is thus obtained in path 88 and then transmitted to algorithm 52.

[0083] Depend on Figure 3 You can see the analysis and evaluation steps. This analysis and evaluation step is designed to be performed on a terminal device, such as a smartphone. The first box 70 represents the cloud, and the second box 72 represents the smartphone.

[0084] In a favorable design, image data collected by the end user is transmitted to the cloud in box 74 and enters the image pipeline in box 76. Figure 2 This image pipeline is related to... Figure 1 The same image pipeline as image pipeline 14 in the image pipeline.

[0085] After the data is pre-processed and analyzed, step 78 checks whether there are enough images. If so, the process proceeds to path 80, and a classification algorithm is executed in box 82. As a result, on the output side of box 82, classified colors are present at box 84. These colors are then transmitted to the smartphone 72 via box 86.

[0086] The smartphone receives data in box 88 and the color classification ends in box 90.

[0087] Conversely, if it is determined in step 78 that there are not enough images, the process proceeds to path 92. In this case, color sorting begins in box 94 by the smartphone, and a photo is taken at point 96.

[0088] The taking of a photo or image is thus triggered or initiated via path 92.

[0089] On the output side of smartphone 72, an image exists in path 98. In box 100, this image is transmitted to the cloud via path 102 and loaded there, so that the execution process in box 74 can be initiated for this image.

[0090] The following describes an exemplary learning algorithm:

[0091] If the algorithm completes training, then as by Figure 4 As can be seen, the algorithm can be viewed at the highest level as a function that assigns a natural number (including 0) from 0 to N (the number of categories) to each input image. The output number represents the different categories, and thus the upper limit of the number is related to the usage or the number of different targets to be classified. In this design according to the invention, these are the 16 different colors (A1-D4) of a tooth color chart.

[0092] Figure 4 An algorithm is shown as a function with an input and an output. This algorithm belongs to a variant of neural networks, the so-called CNN (Convolutional Neural Network). CNNs are neural networks that are mainly used for classifying images, that is to say naming what they see, grouping images according to similarities (picture search) and recognizing objects in a scene. In this way, CNNs are used, for example, to recognize faces, people, street signs, tumors, animals and many other aspects for visualizing data.

[0093] Experiments have shown that CNNs are particularly effective in image recognition and enable deep learning. A deep convolutional architecture known per se, named AlexNet (ImageNet competition, 2012), can be used; at the time, applications of this deep convolutional architecture for autonomous vehicles, robots, drones, the security sector, medical diagnostics were discussed.

[0094] The CNN according to the application does not perceive images as a human being does as it is used. Rather, it is important how the image is supplied to the CNN and how it processes the image.

[0095] The CNN perceives the image more as a volume, that is to say as a three- dimensional object, and not as a planar canvas-like structure that is only measured in width and height. This is because digital color images have a red-blue-green coding (RGB), in which the three colors are mixed in order to produce the color spectrum perceived by humans. The CNN records this image as three separate color layers that are stacked on top of each other.

[0096] That is to say, the CNN receives a normal color image as a rectangular box, the width and height of which are measured by the number of pixels along these dimensions, and the depth of which comprises three layers, one for each letter in RGB. This depth layer is called a channel.

[0097] These numbers are features on the initial, raw perception that is input into the CNN, the purpose of which is to find out which of these numbers is an important signal that helps it to accurately classify the image in a certain class.

[0098] The CNN consists roughly of three different layers, via which the input image is successively propagated via mathematical operations. The number, properties and layout of these layers can be varied depending on the application purpose in order to optimize the result. Figure 5 A possible architecture of a CNN is shown. The following paragraphs specify the different layers in detail.

[0099] Figure 5A schematic diagram showing one CNN convolutional layer. Instead of focusing on pixel by pixel processing, CNN receives a square region of multiple pixels and lets the region pass through a filter. This filter is also a square matrix, which is smaller than the image itself and has the same size as the field (region). The filter is also called kernel and the task of the filter is to find patterns in the pixels. This process is called convolution or convolving.

[0100] The next layer in a CNN has three names: max-pooling, down-sampling and sub-sampling. Figure 6 A schematic diagram showing one exemplary max-pooling layer. The input of the previous layer is fed into the down-sampling layer, which applies this method block-wise, as in the convolution. In this case, in max-pooling, simply the maximum value is taken from one region of an image in a max-pool (see Figure 6 ), the maximum value is filled into a new matrix together with the maximum values from other regions and the rest containing information in the activation map is discarded.

[0101] A lot of information about the smaller values is lost in this step, which motivates the search for alternative methods. But this down-sampling has the advantage of reducing the storage and processing expenditure, which is also due to the loss of information.

[0102] Fully connected layer (Dense Layer)

[0103] The fully connected layer is a "traditional" layer, which is also used in classic neural networks. The fully connected layer consists of a variable number of neurons. The neurons in these layers have a full connection to all outputs of the previous layer, as in normal neural networks. The output of the fully connected layer can thus be calculated with matrix multiplication with subsequent bias-shift.

[0104] Learning process

[0105] The learning process of a CNN is mostly the same as the process of a classic neural network.

[0106] CNNs have so far only been used for recognizing objects in images, but according to the present application, this architecture is used for classifying the color of an object with very high precision.

Claims

1. Method for determining the color of a tooth, wherein - the analysis evaluation device has an iterative learning algorithm using a CNN, which, in a preparatory step, acquires images at different illuminations and from different angles of view based on at least one previously known sample tooth color and analyzes and evaluates the images and learns the correspondence to the correct sample tooth color, - in the analysis evaluation step, - a camera is provided, with which an image of a support body having a previously known sample tooth color together with at least one tooth is taken, - the camera acquires at least two images of the combination of the support body and the tooth to be determined from different angles of view and transmits the images to the analysis evaluation device, and - the analysis evaluation device analyzes and evaluates the acquired images based on the learned correspondence to the correct sample tooth color and outputs the tooth color of the tooth to be determined from the reference value, - the support body has a reference point, and in the analysis evaluation step, the analysis evaluation is only started if the reference point is recognized in the image to be analyzed and evaluated, and in the absence of the reference point, another image is requested, - in the analysis evaluation step, the analysis evaluation device looks for reflections in the tooth segment and / or the support body segment, and only continues the analysis evaluation if only reflections below a predefined threshold are recognized, and when the reflections are above the predefined threshold, another image is requested.

2. The method of claim 1, wherein, The camera is a camera of a terminal device or a scanner.

3. The method of claim 2, wherein, The terminal device is a smartphone.

4. The method of claim 1, wherein, In the preparatory step, the iteration of the learning algorithm is ended when the accuracy of the correspondence to the correct sample tooth color exceeds a threshold value, or when the change of the data provided in the iteration relative to the data of the previous iteration is below a threshold value.

5. The method according to any one of claims 1 to 4, characterized in that, At the end of the preparatory step, the analysis evaluation device or its data is brought into the cloud, and in the analysis evaluation step, the transfer of the at least two images to the cloud is realized.

6. The method according to claim 2 or 3, characterized in that, The terminal device outputs the tooth color of the tooth to be determined.

7. The method according to any one of claims 1 to 4, characterized in that, In the analysis evaluation step, the analysis evaluation device divides the image to be analyzed and evaluated into segments.

8. The method of claim 7, wherein, In the analysis evaluation step, the analysis evaluation device recognizes a segment as a tooth segment if the segment has an area that is essentially tooth-shaped and has a small color and brightness difference similar to a tooth.

9. The method of claim 8, wherein, In the analysis evaluation step, the analysis evaluation device recognizes a segment as a support body segment.

10. The method of claim 8, wherein, In the analysis evaluation step, the analysis evaluation device determines the dominant color of a segment and analyzes and evaluates based on the learned correspondence according to the method of least color difference.

11. The method of claim 9, wherein, In the analysis evaluation step, the analysis evaluation device also analyzes and evaluates based on a comparison of the color and brightness values of the support body segment and the tooth segment.

12. The method of claim 2 or 3, wherein, The terminal device has an ambient light sensor, the output signal of which is transmitted to the analysis evaluation device.

13. The method according to any one of claims 1 to 4, characterized in that, After a predefined number of analysis evaluations, the analysis evaluation device re-performs the preliminary step with the inclusion of the completed analysis evaluations.

14. The method of claim 2 or 3, wherein, The analysis evaluation device has a geometry detection unit which outputs an alarm signal based on the orientation or non-parallelism of the segment boundaries and displays the alarm signal on the screen of the terminal device as an indication of the desired changed orientation of the terminal device.

15. The method according to any one of claims 1 to 4, characterized in that, Known tooth shapes occurring in practice are stored in the analysis evaluation device for determining the tooth segments and / or for improving the geometry detection and are compared with the detected shapes or segment boundaries.

16. The method of any one of claims 1 to 4, wherein, The camera is configured as a camera and the analysis evaluation device comprises an APP of a smartphone which carries out at least a part of the analysis evaluation by the analysis evaluation device.

17. The method of any one of claims 1 to 4, wherein, The camera is configured as a scanner in which a computer is integrated which carries out at least a part of the analysis evaluation by the analysis evaluation device.

18. The method of claim 1, wherein, The sample tooth color is virtually generated in the RGB space.

19. The method of claim 1, wherein, The sample tooth color is virtually generated.

20. The method of claim 1, wherein, The analysis evaluation device outputs the tooth color of the tooth to be determined according to a commonly used tooth shade guide.

21. The method of claim 20, wherein, The analysis evaluation device outputs the tooth color of the tooth to be determined according to A1-D4 of a tooth color shade guide.

22. The method of claim 7, wherein, In the analysis evaluation step, the analysis evaluation device divides the image to be analysis evaluated into segments after recognizing the reference points.

23. The method of claim 10, wherein, In the analysis evaluation step, the analysis evaluation device determines the dominant color of the tooth segment.

24. The method of claim 10, wherein, In the analysis evaluation step, the analysis evaluation device carries out the analysis evaluation according to the method of least color difference taking into account different illuminations.

25. The method of claim 2 or 3, wherein, The camera has an ambient light sensor, the output signal of which is transmitted to the analysis evaluation device.

26. The method of claim 17, wherein, The computer is a small computer.

27. The method of claim 26, wherein, The small computer is a Raspberry Pi.

28. Method for determining a tooth color, wherein - the analysis evaluation device has an iterative learning algorithm using a CNN which, in a preliminary step, acquires images at different illuminations and from different camera angles based on at least one previously known RGB tooth color and analysis evaluates the images and learns a correspondence to the correct RGB tooth color, - wherein, in an analysis evaluation step, - there is a camera with which an image of an aid body having a previously known RGB tooth color together with at least one tooth is taken, - the camera acquires at least two images of the combination of the tooth to be determined and the aid body from different camera angles and transmits the images to the analysis evaluation device, and - the analysis evaluation device analysis evaluates the acquired images based on the learned correspondence to the correct RGB tooth color and outputs the tooth color of the tooth to be determined according to a reference value, - the aid has a reference point and in the analysis step the analysis is only started if the reference point is identified in the image to be analyzed and in the absence of the reference point another image is requested, - in the analysis step the analysis device looks for reflections in the tooth segment and / or the aid segment and only continues the analysis if only reflections below a predefined threshold are identified and when the reflections are above the predefined threshold another image is requested.

29. The method of claim 28, wherein, The camera is a camera of a terminal device or a scanner.

30. The method of claim 29, wherein, The terminal device is a smartphone.

31. The method of claim 28, wherein, The analysis device outputs the tooth color of the tooth to be determined according to a common tooth shade guide.

32. The method of claim 31, wherein, The analysis device outputs the tooth color of the tooth to be determined according to the A1-D4 of a tooth shade guide.

Images