DE4211904C2 - Automatic procedure for creating a list of different types for a liquid sample - Google Patents

Description

The invention relates to an automatic method for creating a list different types for a liquid sample with objects that are shape and / or can be portable and sometimes cover each other or overlap to determine the ecological condition of the sample.  

Such a method, which is carried out manually, is from practice known. It is used to examine activated sludges etc. Biomass, for the determination of water quality and / or for Monitoring biological wastewater treatment systems. To this The purpose is to investigate the "activated sludge" ecosystem a defined space with generally the same properties represents that in the habitat, the non-living biotope, and into the species communities or the community, the biocenosis disintegrates.

The biocenosis is composed of different species, which have very different object densities. Furthermore enough the size of the objects, including algae, fungi, protozoa and metazoa count from 1 µm (bacteria) to several mm (Microorganisms). There are up to 100 and more per sample Object types and each type from one object to several thousand Objects.

Furthermore, not only living objects are found in the sample, but also other particles such as flakes, fibers, hair etc.

With "Art" or "Types" all objects in the following text of the ecosystem, for example both organisms as well as flakes, hair etc.

The purpose of the investigation is to make statements about prehistory, State and likely evolution of the ecosystem to be able to do. It is also about monitoring the Operations and a "health check" of the ecosystem a biological wastewater treatment system. The in the frame Statements obtained from the investigation are used to determine the operational and chemical-physical data of To confirm and / or supplement the system or in the case of a  To initiate further investigations in the event of an accident. The data will continue to be a long-term study of the Biocenosis of the respective plant used.

As part of these investigations, it is necessary to Species list of all objects in a sample (e.g. 50 µl) create and evaluate. The creation of the species list will currently featured by staff with different educational backgrounds a microscope and possibly a CCD camera Video recorder for documentation of the examined samples ver turns.

The classification of the different objects into the different ones Object classes or types are mostly knowledge and experience matter, the methods used depending on the Experience of the person in charge of the investigation is very low are different. Naturally, this is only one incomplete data collection possible, which is only a part represents the biomass to be examined.

The species list created in this way is used for a fort Long-term monitoring (population growth of individual species; Flake formation and age; Complexity of Biocenosis; Oxygen content based on indicator organisms; Saprobia levels as an index for water quality) in a computer entered and to recognize periodic processes already used in the early stage. The computer also carries out a Modeling the biocenosis through to the specific To determine dependencies between the species and so between normal / usual changes in the biocenosis on the one hand and on the other hand, distinguish between deviations caused by the outside to be able to.  

With the examination procedure described so far Not only wastewater treatment plants checked, also natural ecosystems such as lakes, rivers, seas etc. are examined / monitored.

The creation of the species list to be carried out by humans requires not only the limiting time factor but also an undesirable one high error rate. In addition, this is the way Amount of data obtained for a comprehensive overview of the The state of the ecosystem in question is often far too low.

A comprehensive chemical analysis of activated sludge is admittedly possible, but takes even longer than that described at the beginning Determination using a man-made one Species list.

Now it is known to use fluorescence microscopes with camera systems to be used, but the samples are almost exclusively here two-dimensional and dry. Liquid samples with moving Objects can be used with such, especially in research laboratories systems to be found.

Such systems, however, search for desired objects while in the case of the creation of the species list for identification unknown objects whose position in the room can be arbitrary.

In addition, the objects in the liquid to be examined Samples also appear as cluster objects. Among cluster objects we understand several related objects here, e.g. B. several microorganisms can attach to a flake, so that a cluster object arises that consists of inanimate and there are different types of busy objects. The objects such a cluster object can also mutually cover or overlap or even with each other  be intertwined, so that initially a separation of the cluster object in the individual objects is required. As before mentioned, some of the objects in the sample are moving, i.e. H. the objects may change in shape and location. An object type can also be found in various stages of development, so in different sizes. It must also be taken into account that between the dimensions of the smallest and the largest Objects can be almost four orders of magnitude.

Because of the great variety of shapes of living objects not all possible manifestations in the desired process stored in a kind of reference database, because whose number is astronomically high.

With the known image processing methods and systems is not yet the creation of a species list in the desired scope and at the desired speed.

For example, DE 29 03 625 A1 describes an automatic device Blood analysis known. This device uses a digital one Image and pattern recognition system with microscope optics, whereby a sliding table and focus control are used. The blood cells to be analyzed must be spatially separated from one another exist separately in a single-cell layer, whereby only objects be evaluated in a certain size range. When additional criteria determine the shape and color of the objects.

The sample is broken down into individual images, with each individual image the individual blood cells based on self-contained outlines be localized. For such a localized blood cell are geometric parameters such as roundness, crescent shape etc. as well as other morphological parameters such as hemoglobin content etc. determined. Based on these parameters, the blood cell becomes a assigned by several predetermined subpopulations, where this assignment based on the measured and calculated parameters  he follows. For all of the blood cells measured in this way as well statistical data are provided for the individual subpopulations determined, which are then compared with specified standards. Based on this comparison, it is then decided which one Type of anemia the blood examined belongs.

So this system has a recovery expectation regarding the objects to be sorted. Objects in a computer not known form, that is, for example have clustered together or one that is not foreseen Have taken shape are considered a dirt effect discarded. Objects that change shape or even move, which may have clustered together to form larger clusters cannot be analyzed with this device.

DE 33 13 789 A1 describes a microscopic examination device known, in which in the investigative operation nucleated Objects found and based on a number of features, such as for example, color, size and structure. With this device, too, only objects can be classified  for which there is an expectation of recovery. Furthermore, the objects to be examined must pass through a few Identify characteristics that can be determined from the optical data to let.

This method is therefore also unsuitable Objects that can possibly be clustered in locate, identify, and identify a liquid sample classify.

Furthermore, DE 38 36 716 A1 is an interactive method described for the evaluation of cell images, in the technical trained personnel is required in order to operate the To guide and support computers in the evaluation. In order to is this method with the disadvantages already mentioned at the beginning - limiting time factor, high error rate - from manual Procedure afflicted.

A differentiation method is known from WO 91/02330 different particles known. For this purpose, an image of the sample to be examined, whereupon for each the particle in this image is measured as a first parameter which can be the color. Based on this parameter is then a first particle type based on a first threshold identified, whereupon for the identified particles a second parameter different from the first parameter is measured to determine a second inner threshold. This second parameter can be the particle size. Now will the second parameter for all unidentified particles measured to in comparison with the just determined inner Threshold an identification of the remaining particles perform that is greater or less than the first particle type can be differentiated. By determining  the influence of the inner threshold value z. B. aging the sample can be excluded if the relationship of the particles is known to each other with respect to these parameters.

DE 28 23 490 A1 describes a method for the analysis of Organism cells in which the sample to be examined also is optically detected. In a kind of preliminary examination with a lower one The objects that are in a certain wavelength range exceed a threshold. All other particles are excluded from further analysis. The marked cells are then marked with both low resolution, in which the whole cell can be seen, with regard to coarse characteristics such as color, density etc. as well large resolution further analyzed with regard to their fine structures. There is no comparison with reference objects here instead, the particles are based on gray scale histograms classified.

An interactive counting method is known from US Pat. No. 4,932,044 known in which the user cells in one of the discussed Device marked through scanned three-dimensional area. The device then counts the marked cells, taking the area scanned in layers.

A pattern recognition device is known from EP 0 136 718 A2, which the pattern to be identified with a rough and a Fine analysis examined. The rough comparison is used for classification and thus for the selection of reference parameters for the Fine analysis. So the rough analysis narrows the search area, while the fine analysis is used for final identification. The pattern to be examined is two-dimensional Array saved, with analysis for each array element done individually.  

WO 90/10273 describes a method for quality testing known from plants. For this purpose, first a series of identical plants in an interactive learning phase analyzed and thus defined a kind of model plant. The each plant to be examined is then identified by a color classification segmented according to given color classes, d. H. the areas of the plant that belong together in color, such as foliage, Stems and flowers are evaluated separately and included the corresponding data of the model plant for quality control compared. The optically stored image of the The plant is not changed, rather they are different colored regions measured in terms of their area.

Finally, there is a pattern recognition method in US 4,845,765 described for previously known objects which partially overlap each other. In a learning phase, the given Objects processed so that unchangeable segments of the outline as well as the relationships of these segments to each other hierarchically be determined. It is assumed that the individual Objects sufficient due to these unchangeable segments are individualized, so that even with some mutual Coverage of the objects recognized these unchangeable segments and so that the individual objects can be identified. A comparison with reference objects or a processing of the stored image of such a cluster does not take place.

For the reasons already outlined, all of these procedures are for the automatic creation of a list of species for samples with the specified characteristics not suitable.

So it is the object of the present invention, a method for creating to create a list of species for a liquid sample in which  a fast and high data throughput is guaranteed, whereby related objects should also be identifiable.

This object is achieved by a method with the Features of claim 1 solved.

By automatic identification with the help of an identification device errors are largely excluded, as the main source of error and the main time factor in identifying the objects lie. The new process works over it moreover, much faster than the staff on their own. In addition, it is no longer absolutely necessary use qualified specialist staff.

Since the process is broken down into automated individual steps, For example, all objects can be localized before the automatic identification processes run. On the other hand it is also possible if the automatic identification device works very quickly, every localized object to identify immediately.

The process works in such a way that first of all Rough features of the objects are extracted due to image data which is then a narrow one from the existing reference objects Group is selected, with which the then to be identified Objects are compared. So it is, so to speak a knowledge-based process.  

In the process, a cluster object is created automatically is recognized, broken down into sub-objects, which are successively included the available reference objects are compared.

Since the number of possible cluster objects that are in one liquid sample is huge, it wouldn't be possible, all possible cluster objects as reference objects to keep in stock. By breaking the cluster objects into Subobjects becomes the number of reference objects required and consequently the number of comparisons required is strong reduced. This leads to a faster process of the Procedure.

It prefers that automatically identify the next steps:

• - Creation of a modified cluster object by reducing of the cluster object around the one identified Reference object belonging to subobject and
• - Treat the modified cluster object like a localized object.

This measure is advantageous insofar as it is split off of the sub-object is not necessarily a complete main object was split off from the cluster object. It is possible that  the split off sub-object is only part of one on the cluster object hanging single object. By selecting the reference object or main object belonging to the sub-object becomes the number of required decompositions of a cluster object reduced. The cluster object is not just around the sub-object, but around further sub-objects of the to identified sub-object belonging to the main object or reference object reduced. Contains the modified cluster object so possibly fewer sub-objects than if only the previous one addressed sub-object has been subtracted from the cluster object would.

It is also preferred if the automatic detection the presence of a cluster object comprises the steps:

• - Determine a length or maximum extent of the localized Object,
• - Determination of the widths of the transverse to the length localized object and
• - Assessment of the deviations of the widths among each other as well as the position of the latitudes to the longitude.

The advantage of these measures is that they are purely geometric Information, namely the length and the taken transversely to the length Widths can be recognized whether an object or a cluster object is present. The widths of a cluster object are different very different from each other, for example, long, thin objects and spherical objects are connected to each other. Furthermore, there are arbitrary geometric objects in cluster objects Take shape, at least some of the latitude "outside" the Length, d. H. the cluster object is so tortuous that the  Length partially outside the outline of the cluster object runs.

In a development of the method, it is preferred if the step of recognizing and locating the objects automatic transfer of an object into a drawing file and then the automatic detection of self-contained Outlines include.

This measure has the advantage that the transfer of an object in a drawing file first the ones to be processed Data is reduced, not all Image, but only the one or more objects Drawing file can be processed further.

It is further preferred if the optical detection of the sample the automatic recording of image fields of different Magnification includes, with the specimen each magnification above their volume distributed image fields are recorded.

In this advantageous manner, the sample can be added successively various magnification factors are scanned or scanned be so that depending on the size of the to be located Objects the entire volume of the sample with a certain Number of image fields can be completely captured. Since this Scanning is done automatically when browsing manually a liquid sample common error - it certain areas of the sample are "forgotten" - avoided. The sample will also have all the required magnification factors searched. Because of the almost four orders of magnitude sweeping different sizes of each object  this leads to a high volume of data. Through the automated The process ensures that all data is recorded and processed.

It should be mentioned here that under "Image fields" a compilation or summary of "image data" is understood that one play back a specific section of the sample. With "image fields" it is consequently "related" image data.

In this method, it is further preferred if the step of localizing the merging of several image fields, via which extends an object to form an overlay image.

This advantageous and simple measure makes objects, the size of which is such that they cover several image fields extend, yet localized as an object and can be in assigned a reference object to a single comparison process become.

In a development of the method, it is preferred if the step of recognizing and locating the objects or Sub-objects the detection of areas of the same or similar True color, real color pattern or real color areas.

This measure is advantageous in that it is in the image data typical colors for flakes, algae, threads etc. as well as for the Background there. In this way, for example algae that appear green or yellowish recognizing flakes from the always different colored background be distinguished.  

It is further preferred if the comparison between object or search object on the one hand and reference object on the other the basis of three-dimensional virtual vector graphics.

The quasi-optical virtual comparison is compared the pure comparison of extracted rough features is the advantage achieved that features are also included in the identification can be, which is not in the form of, for example Have algorithms laid down. So the reference objects Have very fine structures, on the basis of which easy identification of the search objects is possible while splitting of these very fine structures in characteristics that are a binary decision tree would be based, does not seem possible.

It is also preferred if the step of automatically recognizing a cluster object the disassembly of the cluster object into different areas True colors, real color areas and / or real color contrasts.

This measure is advantageous in that a cluster object always in areas of different colors and color contrasts can split up. This way it is possible to get out of the Image data additional - relating to physical quantities - To gain statements about a cluster object that is a decomposition enable in subobjects.

It is also preferred if the cluster object is in Areas of different movement, geometric shape and / or Size is disassembled.

These criteria also advantageously enable one automatic detection of sub-objects in a cluster object.

The invention is achieved with the help of following description explained in more detail. It shows

Figure 1 shows the device for performing the method in a schematic overall view.

FIG. 2 shows the control and evaluation unit of the device according to FIG. 1 in a basic circuit diagram;

Figure 3 is a with the apparatus of Figure 1 to be examined sample with schematically shown objects..;

FIG. 4 shows a series of image fields of the sample according to FIG. 3 recorded with the image recording device according to FIG. 1;

Fig. 5, the locating means of the control and evaluation unit according to Figure 2, in a schematic block diagram.

FIG. 6 is a more detailed representation of an image field of the sample of Figure 3, with a number of objects.

FIG. 7 shows the identification device of the control and evaluation unit according to FIG. 2 in a schematic block diagram;

FIG. 8 shows one of the cluster objects from the image field according to FIG. 6 in a simplified representation;

FIG. 9 shows the cluster detection device of the identification device according to FIG. 7 in a schematic block diagram; FIG. and

FIG. 10 shows the search device of the identification device according to FIG. 7 in a schematic block diagram.

Fig. 1 a, generally designated 10 shows apparatus for creating a list of species for between a slide 11 and a cover glass 12 contained liquid sample 13. The sample 13 may have objects 14 a, 14 b of different types that are variable in shape and / or location. The device 10 and the method to be carried out with it serve to draw up a list of species and thus to determine the ecological condition of the sample 13 , which can be taken, for example, from a sewage treatment plant or a natural body of water.

There is often a separate one for limnological examination Sampling required for microorganisms such as worms, insects or snails (range mm to cm) and under to be able to search. Such samples are also liquid not on a slide, but in a petri shell or a watch glass.

The sample rooms to be examined are therefore liquid and generally optically transparent. Because of the high variability of the shape of the different objects 14 on the one hand and the fact that the objects 14 are not always isolated, but can overlap, overlap and interweave with one another, on the other hand, the device 10 often has no recoveries compared to the object 14 in the concrete form maintenance, it is so "unknown" to her. The method to be carried out on the device 10 is therefore knowledge-based and self-adapting.

The device 10 has an image recording device 16 , which includes an image recording system 17 and an adjustable optical magnification device 18 . Furthermore, a driving device 19 is indicated, via which the sample 13 can be moved relative to the enlarging device 18 . This method takes place in at least two of the three axes X, Y and Z of a coordinate system indicated at 20 .

In the exemplary embodiment shown, the moving device 19 is a cross table 22 , which can be moved in the X / Y plane and possibly has a lifting device for changing its Z coordinate.

The magnification device 18 comprises a microscope 23 , which can set various magnification factors. By correspondingly focusing the microscope 23 , the Z plane, which is captured by the image recording system 17 designed as a CCD camera 24 , can also be set or changed.

In a limnological examination of microorganisms, a stereo magnifying glass (not shown in the drawing) is used instead of the microscope 23 , to which a CCD camera 24 is also connected. Depending on the size of the objects to be examined, an image recording device without an upstream magnification device can also be used.

Furthermore, a control and evaluation unit 26 is provided, which is connected to the CCD camera 24 via a data line 27 and a control line 28 . In this way, a controlled recording and transmission of image data into the control and evaluation unit 26 takes place . The CCD camera 24 is also connected via a white data line 29 to a mass storage device 31 , which is, for example, a video recorder and is used for video archiving of the examined sample. The mass storage device 31 is likewise connected to the control and evaluation unit 26 via a data line 32 , which can thus query the image data either directly from the CCD camera 24 or from the mass storage device 31 .

A data terminal / terminal 34 is connected to the control and evaluation unit 26 via a connecting line 33 in order to display image data of the sample 13 to a user and to give him the opportunity to intervene in the evaluation process.

Furthermore, two control lines 35 and 36 are indicated by dashed lines, via which the control and evaluation unit 26 controls the cross table 22 and the microscope 23 .

With the device described so far, the sample 13 is scanned (scanned) one after the other in the X, Y and Z directions, so that three-dimensional image data are available at the end of the scanning process. As mentioned, some of the objects 14 are movable, so that additional information on the time axis can also be obtained by scanning again. The quality of the images is essentially determined by the lens of the microscope 23 and by the resolution / digitization of the captured image. The adjustable magnification of the microscope 23 enables objects 14 of very different sizes to be detected. The image data represent true color images. Likewise, it is possible to subject the sample 13 to a fluorescence examination by irradiating it with a light source not shown in FIG. 1, this excitation light being filtered out again by a corresponding filter in the microscope 23 .

The captured image data is evaluated in the control and evaluation unit 26 shown in more detail in FIG. 2. Via the data lines 27 and 32 , image data are loaded either directly from the CCD camera 24 or from the mass storage device 31 into an image data storage device 38 . Via its output line 39 , the image data reach a localization device 40 , which "perceives" the objects contained in the images in a manner to be described in more detail, ie recognizes and localizes them. The objects located in this way are transferred via an object data line 41 to an object data memory 42 , from which they arrive at an identification device 44 via its output line 43 .

The identification device identifies the localized object present in the object data memory 42 , ie it determines its type by associating the object 14 as a member with a corresponding type in a reference object memory to be described. The identification device 44 is connected via an object output line 45 to a counting device 46 , which counts the identified objects and enters a type list indicated at 48 via a transfer line 47 .

Furthermore, an evaluation device 49 is indicated in FIG. 2, which carries out an automatic evaluation or interpretation of the data entered in the species list 48 using target-oriented methods from ecology, biology and / or etology. On the basis of the interpretation of the species list 48, the evaluation device 49 provides statements about the ecological condition and, if necessary, a prognosis about the development of the sample 13 under investigation.

The device 10 described so far and the method to be carried out on it are now to be explained in more detail with reference to FIGS. 3 to 10:

In Fig. 3, the sample 13 of FIG. 1 is shown in an enlarged scale. By imaginary dividing lines 50 , the sample 13 is divided into a grid of image fields 51 , which are scanned one after the other by the image recording device 16 from FIG. 1. For this purpose, the cross table 22 is moved to a specific X / Y coordinate and then image fields 51 of a stack 52 lying one above the other in the Z direction are captured by the CCD camera 24 , digitized and transferred to the data lines 27 , 29 as image data. The size of the individual image fields 51 depends on the magnification factor to which the microscope 23 is set. The distance between two superimposed image fields 51 a, 51 d is determined by the depth of field of the microscope; namely, the smaller its depth of field, the more image fields 51 in a stack 52 have to be captured in order to scan the entire thickness of the sample 13 .

When a stack 52 has been "processed" in this way, the cross table 22 moves either in the X or Y direction by the width / length of an image field, and the successive movement in the Z direction starts again. In this way, the sample 13 is scanned in a meandering manner, so to speak. Of course, it is also possible not to arrange the image fields 51 in different Z planes in the form of a stack 52 , but to offset them by a small amount in each case, so that a zigzag travel line is created over all. Compared to meandering scanning, fewer travel steps then have to be undertaken in order to scan the entire sample 13 .

Of course, a sample 13 is scanned again for different magnification factors of the microscope 23 . In this way, first the large objects 14 with a low magnification factor and then with increasing magnification ever smaller objects 14 are recognized, localized and identified so that they can be entered in the species list 48 .

In Fig. 3 it can be seen that the objects 14 can take any positions in space, so that they can extend over several image fields 51 .

As can be seen in FIG. 4, an object 14 , which extends in a stack 52 over a plurality of image fields 51 c, 51 d and 51 e lying one above the other, produces a different pattern 14 ′, 14 ′ ′ and 14 in each image field 51 '''. In order to be able to identify the object 14 , it is first necessary to decide how many image fields 51 an object 14 extends over. In the example shown in FIG. 4, the object 14 extends over the image fields 51 c- 51 e. Due to the depth of field of the microscope 23 , certain sections of the object 14 are depicted sharply in the plane of the image field 51 , as indicated by solid lines in FIG. 4, while other sections of the object 14 can only be seen as blurred. This is indicated in Fig. 4 by dashed lines.

In order to be able to position the individual image fields one above the other in the correct position, the image fields 51 must first be rotated and shifted in the X, Y and / or Z direction in order to take into account the movement of the object 14 in question. In order to recognize whether different image fields 51 a, 51 e are an object 14 that has moved further between the exposures of the successive image fields 51 , or whether different objects 14 , 14 b are being viewed, additional information can be used that come from image data that were obtained with a lower magnification factor of the image recording system. With a lower magnification factor, that is, with a different depth of field, it is easier to distinguish whether there are two objects 14 b, 14 lying one above the other or a single, moving object 14 . With this superimposition of the individual image fields 51 , information falls off, so to speak, as to whether it is a moving object 14 .

Since 51 c 51 e passes a period of time between shots of the images, the object 14 may have meanwhile moved, but that can be detected on the basis of the difference between sharp contours and blurred contours in different Z-levels and balanced. A further stack image of the object 14 b is indicated at 53 b.

If the individual image fields 51 c- 51 e are positioned one above the other in the correct position, a stack image 53 is obtained which shows a kind of contour image 54 of the object 14 . This stack image 53 means that the objects 14 are now in the form of three-dimensional image data, so that they can be recognized in the localization device 40 and identified in the automatic identification device 44 . To do this, they are stored as vector graphic objects.

Of course, the objects not only extend through a plurality of image fields 51 in the Z direction, it is quite possible, in the same way, as shown in FIG. 4 for the Z direction, also in the X and / or Y direction Recognize an object 14 .

In Fig. 5 the competent locator 40 is shown in detail profiled. The data of the individual image fields 51 that are present in succession on the output line 39 first reach an outline detector 56 , which searches the image data for self-contained outlines and / or areas of the same color, and thus to the sharp (solid) contours in the image fields 51 a- 51 b from FIG. 4 comes. This takes advantage of the fact that the objects 14 are always different in color from the background, so that they differ from the background not only by the contrast but also by the coloring. In addition, the outline detector 56 detects the blurred outlines of the objects 14 .

For this purpose, the outline detector 56 places an overlay window over the respective objects as follows: First, point by point of each image field 51 is queried as to whether its true color belongs to the background color area or stands out from it. If a picture element (pixel) is found that does not belong to the background, so that it is part of an object 14 , then this point is covered by a virtual square window. The edges of the window are now iteratively shifted away from each other in all four coordinate directions until the edges are only over pixels that can be assigned to the background color range. Since the starting pixel was an object pixel, this ensures that at least one object is delimited in the overlay window generated in this way.

The object 14 thus delimited is copied into a partial image, so that the data to be processed further have been significantly reduced. After copying, the object (the objects) recognized in the original image data set is "deleted" by converting the recognized object pixels into background pixels. The original image data thus only contain "non-perceived" objects that are recognized and copied in the same way.

While moving the overlay window edges, the outline detector 56 has simultaneously marked each object pixel as such and has additionally identified pixels belonging to the outline of the respective object 14 in the following way: Each pixel which is connected to a pixel on at least one of its four sides bordered with background color, "belongs" to the outline of the object, while pixels surrounded on all sides by object pixels are "inside" the object. The method just described also ensures that each edge of the overlay window adjoins the enclosed object at least at one point.

After the individual objects in the different image fields 51 have been recognized in this way, the reduced data arrive in a stack image overlay 57 which , as already described with reference to FIG. 4, superimposes the individual image fields 51 so that the blurred and the sharp contours differ Image fields 51 match. The individual stack images 51 a- 51 e, which are aligned with one another in this way, are stored in the object locator 58 as three-dimensional vector graphic representations. From here, the data arrive in a motion detector 59 which , on the basis of the displacement of the individual image fields 51 that was necessary in order to arrive at an object 14 , determines whether it is a moving object. Via the output lines 60 and 61 , the data of the stack image 53 and the movement information are fed to a logic circuit 62 , which gives this data via the object data line 41 to the object data memory 42 .

In FIG. 6, the projection of a stack image 53 generated by the locator 40 is shown fragmentary, such as is typically found an activated sludge for a sample 13. The stack image 53 shows a small flake 64 , a protozoan 65 and a filamentous bacteria 66 with additional bacterial growth 67 . Furthermore, a large flake 68 can be seen on which another thread bacterium 69 is seated. Furthermore, Fig. 6 has a colony of three bell animals 70 a, 70 b, 70 c, which have a common stem 71 , which sits with its base 72 on the large flake 68 . While the objects 64 and 65 are individual objects that can be easily identified in the identification device 44 , the objects 66 , 67 and 68 , 69 , 70 , 71 represent a cluster object 73 a, 73. The number of possible cluster objects is astronomically high that it is not possible to create reference data objects for these cluster objects. The identification device 44 must therefore be able to process any cluster object 73 in such a way that it can be clearly identified with a limited number of predetermined reference data. It should also be borne in mind here that the sub-objects of a cluster object 73 can at least partially overlap or overlap one another and may also be interwoven with one another.

In Fig. 7 it is shown that the identification device 44 comprises for this reason, a cluster decomposing means 74 as well as an object detection device 75. The data of a localized object is sent via the output line 43 to a cluster recognition device 76 , which checks each newly pending object to determine whether it is an individual object or a cluster object 73 . How this happens will be explained later.

If the object pending on the output line 43 is a single object, the cluster recognition device 76 transmits the data via its output line 77 to a comparison memory 78 of the object recognition device 75 . If, on the other hand, it is a cluster object 73 , the data is loaded into a cluster memory 80 by the cluster recognition device 76 via its output line 79 . From the cluster memory 80 , the data passes via a cluster data line 81 to a sub-object recognition device 82 , which breaks down a cluster object 73 into a number of sub-objects. This breakdown into sub-objects is based on further data that, for example, make statements about the geometric shape of certain areas of the cluster. Furthermore, the colors of the cluster areas and / or the different contrasts are used to break down the cluster object 73 shown in FIG. 6 into individual subobjects. Returning to FIG. 6, it can be seen that, for example, the bell animals 70 and their stems 71 are hatched differently, as a result of which a different color is indicated. The sub-object recognition device 82 now splits off the bell animal 70 a from the cluster 73, for example, and transmits the data of the bell animal 70 a via a sub-object data line 83 to the comparison memory 78 .

The comparison memory 78 now contains either the data of a single object 14 or the data of a sub-object. This data, referred to as “search object”, reaches a search device 85 via a search object data line 84 , which in turn is connected to a reference data memory 87 via a reference object data line 86 . The reference data memory 87 contains a vector, three-dimensional representation of all objects 14 that can occur in a sample 13 .

Since many of the objects 14 are changeable in shape, that is to say have a different geometric shape, for example, when they are at rest or moving, this must be taken into account in the reference object data. However, the number of possible shapes that a single movable object 14 can take is so great that they cannot all be predicted and stored. For this reason, the reference data memory 87 is constructed in such a way that the reference objects are stored in one or more basic geometric shapes. In addition to this basic form, degrees of freedom are stored, which take into account the possible and permissible movements and changes in shape of the respective object. When comparing the search object and the various reference objects, the reference objects are now changed with respect to the search object (within the scope of their permissible degrees of freedom) until they either "match" the search object or until it is found that no match can be achieved. This comparison is a virtual optical process in which, based on a knowledge-based method (the permissible degrees of freedom of a reference object are saved), the vector graphic representation of an object / sub-object is compared with a two-dimensional projection of a virtual three-dimensional vector surface graphic model of the reference object to be compared in each case.

The search device 85 therefore now compares the data which are present on the search object data line 84 with the existing data of the reference data memory 87 until the reference object which matches the search object is found. When splitting off in the sub-object recognition device 82 , it can now happen that not an entire object 14 is split off from the cluster 73 , but only a partial object. In the example of Fig. 6 is not cleaved from stem 71 and a bell portion 70 a existing vorticella one, but only the part of object 70 a.

In the reference data memory 87 , however, there is a main object 14 for sub-object 70 a, consisting of sub-object 70 a and sub-object 71 a. This main object is now passed via a main object data line 88 to a virtual optical difference generator 89 , which is also connected to the cluster memory 80 via a line 90 . The difference forming unit 89 now modifies the cluster object originally located in the cluster memory 80 in such a way that it is reduced by the already recognized main object.

The modified cluster object is returned via line 91 to the cluster recognition device 76 , where the image data now pending are either recognized again as a cluster and are written into the cluster memory 80 , or are recognized as a single object and are thus given directly to the comparison memory 78 .

Of course, each recognized main object is reported to the counter 46 via the object output line 45 .

It should be mentioned here that the reference objects present in the reference data memory 87 are expanded on a case-by-case basis by further newly occurring or newly determined objects, the device 10 is therefore self-adapting.

Referring to Fig. 8 will now be described, according to which criteria the cluster recognizer 76 determines whether it is in the forthcoming on the data output line 43 to a single object 14 or a cluster object 73rd

For the sake of clarity, the bell animals 70 b and 70 c have been omitted from FIG. 6 in FIG. 8. The colors and the contrast of the individual elements of the cluster 73 'from FIG. 8 also play no role in this example for the detection of whether it is a cluster. The cluster object 73 is represented in FIG. 8 only by its outline indicated at 92 . This outline of the cluster object 73 has already been determined with the aid of the outline detector 56 - see FIG. 5.

The cluster recognition device 76 now places a radius around the cluster object and thus determines its length, indicated at 93 , its maximum extent in the X / Y plane. This is done, for example, by determining the distance between all coordinate pairs (X / Y) and then recording the maximum distance as length 93 in the image data.

Next, the width of the cluster object 73 or the object 14 is determined perpendicular to the length 93 . This is also done by determining the distance between the corresponding coordinates, some widths are indicated at 94 a, 94 b in Fig. 8. It can be seen that the cluster object 73 partially has widths 94 c that do not have an intersection with the length 93 . This is an indication of an irregular shape of the object 14 or the cluster object 73 . According to criteria to be determined in each case, the cluster detection device 76 determines from the variation of the different widths 94 a- 94 c and from the fact how often the length 93 lies "outside" of the object 14 or the cluster object 73 , whether it is the present data set is a cluster object 73 .

The cluster recognition device 76 is shown in more detail in FIG. 9. The object data of a localized object pending on the output line 43 are fed to individual assemblies of the cluster detection device 76 via an inner data branch 95 . Via a switch 96 , either the output line 43 , which comes from the object data memory 42 , or the line 91 , which contains information from a modified cluster object 73 , is switched to the internal data branch 95 .

These data arrive in a device 97 for determining the length, a device 98 for determining the width and a qualification device 99 which, for. B. determined from the ratio of length to width, whether the object 14 to be evaluated is a single object 14 or a cluster object 73 . The operation of the devices 97 , 98 and the qualification device 99 has already been explained with reference to FIG. 8.

The qualification device 99 outputs the pending data either to the output line 77 if it is a single object 14 , or to the output line 79 if it is a cluster object 73 .

The cluster recognition device 76 furthermore has a device 100 which recognizes on the basis of movement processes in partial areas of a cluster and / or of contiguous areas of the same true colors whether the localized object 14 is a cluster object 73 .

Finally, it will be explained how the search device 85 compares the search objects pending on the search object data line 84 with the reference data of the reference data memory 87 accessible on the reference data line 86 . This is done using FIG. 10.

For this purpose, the search device 85 has a virtually optically operating comparator device 101 , which sequentially queries the available data of the reference data memory 87 and compares it with the search object data on the search object data line 84 . So that not all reference objects have to be screened for each search object, the search device 85 comprises, for example, four evaluation units 102 , 103 , 104 and 105 , which qualify the search objects according to color, movement, size and geometric shape or changeability in shape. This information leads to decision criteria which enable the comparator device 101 to actually compare only a small set of the total available reference objects with the respective search object.

If, for example, the longest dimension of the search object is less than two μm, only the bacteria stored in the reference data memory 87 are compared with the search object. If, on the other hand, the search object has an unusual, rare coloring, only the reference objects identified with the corresponding color code are compared with the search object. Corresponding considerations can also be made for movement and geometric shape.

Depending on which of the four criteria color, movement, size and geometric shape that is shown here as an example proves to be the most relevant, the comparator device 101 searches certain areas of the reference data memory 87 . In this way, a much faster identification of the object and thus a cluster 73 in a recursive manner can be carried out.

The evaluation devices 102-105 represent a feature set comparator , so to speak, which performs a type of morphological classification of the objects on the basis of features extracted from the image data. Because of the great variety of shapes of the individual objects of the same type, such a morphological classification is only possible to a certain extent due to rough features. Then, as already described above, a quasi-visual comparison takes place in the comparator device 101 , a comparison based on virtual three-dimensional vector graphics. The vectors of the search and reference object are compared with one another in terms of magnitude (length) and solid angle (orientation in space). The reference objects are designed to change their shape, so they are changed with regard to the respective search object until they "fit". In this way, all features that cannot be easily described can still be taken into account, since they are implemented in the finely specific structure of the reference object. A complete object identification based on descriptive features is also not possible because the different types differ from each other in part by sets of features that are difficult to detect in algorithms, be it at the hardware or software level.

If the search object has been recognized as part of a main object, the corresponding main object is loaded by the comparator 101 into a main object memory 106 which is connected to the difference former 89 via the main object data line 88 .

For the sake of completeness, it should be mentioned that the modules indicated in FIGS. 2, 4, 5, 7, 9 and 10 can be implemented in any combination wholly or in part either by pure hardware circuits / microprogram control units, arithmetic circuits or program-controlled data processing units. Furthermore, it goes without saying that the individual memories can either be separate memories or can be partially overlapping areas of a single memory.

Claims (9)

1. Automatic method for creating a list ( 48 ) of different types for a liquid sample ( 13 ) with objects ( 14 ), which can be shape and / or location-changing and partly overlap or overlap, to determine the ecological state of the sample ( 13 ), the sample ( 13 ) being optically recorded at least two-dimensionally as a sequence of image data and the image data being stored and the following steps then being carried out automatically:

• Recognition of the presence of the objects ( 14 ) and localization of the objects ( 14 ) in the captured image data by means of a localization device ( 40 ),
• - Identifying the localized objects ( 14 ) by means of an identification device ( 44 ) which comprises a cluster decomposition device ( 74 ) and an object recognition device ( 75 ), wherein
• - The cluster decomposition device ( 74 ) automatically detects the presence of a cluster object ( 73 ) consisting of a plurality of connected objects ( 14 ) by means of a cluster detection device ( 76 ) and an automatic splitting off of a sub-object from the cluster object ( 73 ) by means of a difference-forming unit ( 89 ) , and where
• - In the object recognition device ( 75 ), the data of the objects ( 14 ) and / or sub-objects are compared with data from reference objects by means of a search device ( 85 ) and thereby identified,
• - The comparison is carried out in such a way that at least one rough feature is determined from the image data of objects ( 14 ) or subobjects to be identified by at least one evaluation unit ( 102, 103, 104, 105 ), on the basis of which then only a certain group of reference objects Reference objects are selected for comparison, which is carried out using vector graphics,
• - Counting the identified objects ( 14 ) according to the different types and
• - Enter the counted objects ( 14 ) in the list ( 48 ) of different types.

2. The method according to claim 1 with the further steps:

• - Generating a modified cluster object ( 73 ) by reducing the cluster object ( 73 ) by the reference object belonging to an identifying sub-object and
• - Treating the modified cluster object ( 73 ) as a localized object ( 14 ).

3. The method according to claim 1, wherein the automatic detection of the presence of a cluster object ( 73 ) comprises the steps:

• Determining a length ( 93 ) or maximum extent of the localized object ( 14 ),
• - Determining transverse to the length ( 93 ) widths ( 94 ) of the localized object ( 14 ) and
• - Assess the deviations of the widths ( 94 a, 94 b, 94 c) with each other and the position of the widths ( 94 ) to the length ( 93 ).

4. The method according to any one of claims 1 to 3, wherein the step of recognizing and locating the objects ( 14 ) comprises the automatic transfer of an object ( 14 ) into a partial image and then the automatic detection of self-contained outlines ( 92 ). 5. The method according to any one of claims 1 to 4, wherein the optical detection of the sample ( 13 ) comprises the automatic recording of image fields ( 51 ) of different magnifications, with the sample ( 13 ) each image over its volume distributed image fields ( 51 ) will. 6. The method according to claims 4 and 5, wherein the localizing step comprises combining a plurality of image fields ( 51 ) over which an object ( 14 ) extends to form an overlay image ( 53 ). 7. The method according to any one of claims 1 to 6, wherein the step of locating the objects ( 14 ) or sub-objects comprises the detection of areas of the same or similar real color, real color pattern or real color areas. 8. The method according to any one of claims 1 to 7, wherein the comparison between object ( 14 ) or sub-object on the one hand and reference object on the other hand is based on three-dimensional virtual vector graphics. 9. The method according to any one of claims 1 to 8, wherein the cluster object ( 73 ) is broken down into areas of different movement, geometric shape and size.