EP1254415A1 - Dispositif, support d'informations et procede pour trouver des objets presentant une grande similitude par rapport a un objet predetermine - Google Patents

Dispositif, support d'informations et procede pour trouver des objets presentant une grande similitude par rapport a un objet predetermine

Info

Publication number
EP1254415A1
EP1254415A1 EP01911439A EP01911439A EP1254415A1 EP 1254415 A1 EP1254415 A1 EP 1254415A1 EP 01911439 A EP01911439 A EP 01911439A EP 01911439 A EP01911439 A EP 01911439A EP 1254415 A1 EP1254415 A1 EP 1254415A1
Authority
EP
European Patent Office
Prior art keywords
objects
value
properties
property
values
Prior art date
Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
Withdrawn
Application number
EP01911439A
Other languages
German (de)
English (en)
Inventor
Ulrich Prof. Dr. Güntzer
Wolf-Tilo Balke
Werner Prof. Dr. Kiessling
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Individual
Original Assignee
Individual
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by Individual filed Critical Individual
Priority to EP01911439A priority Critical patent/EP1254415A1/fr
Publication of EP1254415A1 publication Critical patent/EP1254415A1/fr
Withdrawn legal-status Critical Current

Links

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
    • G06F16/90—Details of database functions independent of the retrieved data types
    • G06F16/903—Querying
    • G06F16/90335—Query processing
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
    • G06F16/50—Information retrieval; Database structures therefor; File system structures therefor of still image data
    • G06F16/58—Retrieval characterised by using metadata, e.g. metadata not derived from the content or metadata generated manually
    • G06F16/583—Retrieval characterised by using metadata, e.g. metadata not derived from the content or metadata generated manually using metadata automatically derived from the content
    • G06F16/5838—Retrieval characterised by using metadata, e.g. metadata not derived from the content or metadata generated manually using metadata automatically derived from the content using colour
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
    • G06F16/50—Information retrieval; Database structures therefor; File system structures therefor of still image data
    • G06F16/58—Retrieval characterised by using metadata, e.g. metadata not derived from the content or metadata generated manually
    • G06F16/583—Retrieval characterised by using metadata, e.g. metadata not derived from the content or metadata generated manually using metadata automatically derived from the content
    • G06F16/5854—Retrieval characterised by using metadata, e.g. metadata not derived from the content or metadata generated manually using metadata automatically derived from the content using shape and object relationship
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
    • G06F16/50—Information retrieval; Database structures therefor; File system structures therefor of still image data
    • G06F16/58—Retrieval characterised by using metadata, e.g. metadata not derived from the content or metadata generated manually
    • G06F16/583—Retrieval characterised by using metadata, e.g. metadata not derived from the content or metadata generated manually using metadata automatically derived from the content
    • G06F16/5862—Retrieval characterised by using metadata, e.g. metadata not derived from the content or metadata generated manually using metadata automatically derived from the content using texture

Definitions

  • the invention relates to methods according to the preamble of patent claim 1, 2, 8, 12, 13, a device for carrying out the methods and a storage medium which can be read by a computer and on which the methods are stored.
  • a method for determining objects with great similarity to a given object is used, for example, when searching in information systems.
  • the handling of multimedia data such as images, video or audio data in information systems, in which objects are sought which correspond to a given object with the greatest possible similarity, require particularly efficient search methods due to the complexity of the data and the large amounts of data.
  • a search evaluation in relation to the similarity to a given object a set of objects is not found which exactly corresponds to the given object, but rather a set of objects is determined which correspond to a more or less high degree of similarity to the given object ,
  • a corresponding method is known, for example, from Fagm, "Combining Fuzzy Information from Multiple Systems", 15th ACM Symposium on Pipnipiples of Database Systems, pp. 216 to 226, ACM 1996.
  • a predetermined number of objects that a predetermined one Have the number of properties, the number k of objects selected, which a given object, which is referred to below as a sample object, with predetermined egg best of all.
  • the database in which the objects with the properties are stored is searched and a data list is determined for each property.
  • the data lists are sorted according to the descending values of the properties.
  • the data lists are also referred to as atomic output streams.
  • the sample object is defined by values in given properties.
  • a combination function is specified with which the values of the properties of the objects to be compared are evaluated in order to obtain information about the most similar objects.
  • the calculation of the combination function with the properties yields a value number for each object, which is also referred to below as an aggregated score.
  • the task of the method is now to determine the k objects with the highest aggregated scores for the given object.
  • the search is performed using the data lists for the properties according to the following procedure.
  • the aggregated scores S (x) F (si (x), ..., s n (x)) are determined for each object x, with s (x) the value of the property I des Object x and with F the combination function is designated and the variable 1 is a natural number that fulfills the following condition: l ⁇ i ⁇ n.
  • the k objects that have the largest aggregated scores are then selected and output as a result.
  • the Fagm method is relatively time-consuming since a large number of objects must be selected and all objects must be directly accessed to the previously unknown properties of the objects. Direct access is relatively time-consuming and costly, especially with heterogeneous information systems.
  • the object of the invention is to provide a more efficient and faster method for determining objects that best ah- no to a given object.
  • An advantage of the invention according to claim 1 is that the value number of the objects is compared with a comparison number and the number of objects to be considered is thereby restricted in a simple and efficient manner
  • An advantage of the invention according to claim 2 is that only the objects are considered whose values for the properties under consideration are above a determined limit value. This also effectively limits the number of objects to be checked.
  • a particularly efficient method is achieved in that the comparison number is calculated with the combination function using the smallest values of the properties of the selected objects.
  • a further improvement of the methods is achieved in that the values of the properties of a selected object that have not yet been selected are estimated by the smallest values that have already been selected for the corresponding properties.
  • FIG. 4 shows a data list for the texture property
  • FIG. 5 shows a data list for the Faroe property
  • FIG. 10 shows another data list for the color property
  • FIG. 14 shows a third data list for the texture property
  • FIG. 15 shows a third data list for the color property
  • FIG. 22 shows a flowchart for a fourth method
  • FIG. 23 shows a further data list for the texture property
  • the information system is preferably implemented in the form of a computer system, the methods for determining the most similar objects preferably running automatically.
  • the information system has an input / output device 1, which is preferably designed as a graphical user interface.
  • the input / output device 1 is connected to a search engine 2 m.
  • the search engine 2 accesses the database 3, which has a visual extender, a text extender and an attribute-based search system.
  • the visual extender, the text extender and the att ⁇ but-based search system represent program blocks in which programs for color recognition, texture recognition, text recognition or internet research are stored
  • a selection device 4 is provided, which is connected to a data memory 6 and to the database 3 m.
  • the selection device 4 is connected to a formatting device 5 m, which in turn is connected to the input / output device 1.
  • the information system according to FIG. 1 functions as follows:
  • the object to which similar objects are sought is entered via the em / output device 1, and that in is referred to as a sample object.
  • the object is called a sample object because it serves as a search pattern for comparison with the objects to be checked.
  • the properties of the object and the combination function are entered with which the properties of the objects are evaluated during the comparison.
  • the object is not limited to pictorial patterns, but can represent any type of form or information.
  • the search engine 2 determines a data list from the database using the program blocks of visual extenders, text extenders and att ⁇ but-based search system for each property that was defined as the search criterion for the given object (sample object).
  • the specified program blocks are only examples. A person skilled in the art will use the programs which are best suited for the search for the method according to the invention.
  • the objects are listed sorted by the value of the property.
  • the data lists and the specified combination function F are output to the selection device 4 and stored in the data memory 6.
  • the selection device 4 uses the data lists and the combination function F to determine the predetermined number of objects which most closely correspond to the specified object (sample object).
  • the predetermined number of best objects is passed on by the selection device 4 to the formatting device 5, which processes them according to a predetermined format and outputs them via the emitting / output device 1.
  • the individual function blocks of FIG. 1 can also be implemented in the form of programs and / or electronic circuits.
  • FIG. 2 shows an example of data lists 12, 13 for the properties 1 to n.
  • a first data list 12 an identifier OID for the objects is in a first column, and the rank of the object within the column in a second column Data list and in a third column the value of the property of the object.
  • the objects are sorted in the data lists of the individual properties in such a way that the object with the largest value is in first place and the other objects are distributed among the other ranks according to the descending value.
  • FIG. 3 shows a flowchart of a first algorithm with which a predetermined number of objects are selected from a predetermined number of objects, which best fit a predetermined object (sample object) with predetermined properties without having to search the entire database.
  • This method largely avoids direct access to the data in the database, so that the method can be carried out quickly and inexpensively.
  • n properties and a combination function F are entered into the input / output device 1 for the predetermined object, which is referred to below as a sample object.
  • the properties and the combination function can be freely defined.
  • the properties are preferably determined depending on the sample object in such a way that the properties of the sample object are selected that best describe the sample object.
  • the combination function F is also preferably determined in such a way that the more defining properties of the sample object are rated higher than the less defining properties.
  • the search engine 2 determines from the database 3 at program point 21 a data list corresponding to FIG. 2 for each property entered, in which the objects are listed sorted by descending value.
  • the selection device 4 selects the object with the greatest value of the property, which has not yet been selected for this property a first data list and stores the value of the property with the identifier OID of the object for the property under consideration in a result list in the data memory 6.
  • the selection device 4 then checks at program point 23 whether all properties to be considered of the object selected at program point 22 have already been stored in the result list. If this is not the case, the selection device 4 determines all unknown properties of the selected object at program point 24 by direct access to the database 3. The properties of the selected object determined from the database 3 are likewise stored in the result list.
  • the selection device 4 calculates, at program point 25, a value number S (aggregated score) for the selected object o using the following formula:
  • the combination function F consists, for example, of the arithmetic mean of the values of all the properties of the sample object considered, if these characterize the sample object with the same strength.
  • the number of values of the effect is also carried in the result list in the data memory 6.
  • the selection device 4 selects the object o tOD from the result list in the data memory 6, which has the largest value number, from the program point 26.
  • the selection device 4 compares whether the value number of the object with the maximum value number, which is stored in the data memory 6 m of the result list, is greater than or equal to the comparison number V.
  • the selection device 4 outputs this object at program point 29 as an object with the greatest similarity to the predefined object via the format device 5.
  • the selection device 4 checks at program point 30 whether the predetermined number k of best objects has been output. If this is the case, the program ends. If this is not the case, the program branches back to program point 22 and the program is run through again.
  • the sequence of the first algorithm according to FIG. 3 is explained in more detail below using a data example
  • the properties of the image that are used for the search are the texture and the color red of the given image (sample object).
  • the arithmetic mean of the two properties is used as the combination function F, since both the color and the texture shape the pattern object equally:
  • FIG. 5 show the data lists which, in this example, are determined by the search engine 2 from the database 3 and are supplied to the selection device 4.
  • the data list Si in FIG. 4 represents a list of objects which are sorted according to the texture property with a descending value.
  • the data list s 2 of FIG. 5 represents a list of objects which are sorted according to the color property with a descending value.
  • the first, second, third, fourth, fifth, sixth etc. object is identified with the identifier OID o x , o 2 , o 3 , o 4 , o 5 , o s etc.
  • the color to be compared is the color red and the texture to be compared is a defined hatching or patterning.
  • the object ⁇ is selected in accordance with program point 22.
  • the query at program point 23 shows that the object ⁇ m is not known to the first three objects considered in the second data list s 2 . Consequently, after program point 24, the value of the color property for the object o- * is obtained via direct access to the database 3. determined.
  • This is also carried out analogously for the objects o 2 , o 3 , o 4 , o 5 , o s m.
  • the values of the missing properties are determined in each case by direct access to the database 3.
  • the values of the objects that are determined during direct access from the database are shown in Fig. 6.
  • the access list is stored in the data memory 6 of the selection device 4.
  • the values of the properties are saved for the first, fourth, second and fifth objects o l7 o 4 , o 2 and o 5 m of the result list.
  • the value numbers are calculated according to program point 25 from the values of the properties and are stored in the result list in accordance with FIG. 7 in data memory 6.
  • the query at program point 28 Before the evaluation of the fifth object o5, the query at program point 28 always showed that the value number of the object S (o cop ) with the maximum value number (aggregated score), which is stored in the result list, is smaller than the comparison number V. The program always branched back to item 22.
  • the object o 4 is selected at program point 26 as an object with a maximum value number (aggregated score), the value number having the value 0.91.
  • the comparison number V is then determined according to program item 27:
  • V F (s ⁇ (rL (z-), ..., s n (r n (z n ))),
  • the program branches to program point 29 and the fourth object o4 is output as the object that best matches the specified object.
  • the program branches to program point 29 and the fourth object o4 is output as the object that best matches the specified object.
  • a second algorithm for determining similar objects is shown using a flow chart in FIG. 8. The second algorithm enables a particularly efficient method for determining a predetermined number k of objects that best match a predetermined object.
  • n predeterminable properties and a predefinable combination function F are entered via the input / output device 1 for a sample object for which similar objects are sought.
  • the sample object, the n properties and the combination function F correspond to those of the first algorithm according to FIG. 3.
  • the search engine 2 determines a data list from the database 3 at program point 32 for the properties texture and color, which are shown in FIGS. 9 and 10.
  • the objects are listed in the data lists sorted by descending value and the data lists are supplied to the selection device 4.
  • the selection device 4 in each case selects the two objects with the highest values from the two data lists and stores the identifier of the objects with the values for the properties in the data memory 6 in a result list.
  • another number p of objects can also be selected. The optimum number p is determined by the person skilled in the art depending on the application.
  • the selection device 4 then calculates an indicator for each data list, the indicator designating the gradient with which the value of the properties falls over the number of objects. For this, only the objects that are stored in the result list are taken into account.
  • an indicator Ii can be calculated as follows for each data list that contains more than p elements:
  • the selection device 4 checks at program point 34 whether all properties of the objects whose identifications are stored in the result list are known. If this is the case, the comparison number V is then calculated at program point 35 using the following formula:
  • V F (s ⁇ (r x (zi)), ..., s n (r n (z n )),
  • the selection device at program point 36 then calculates the value numbers S (aggregated score) for the objects o in the result list using the following formula:
  • the selection device 4 compares the objects which are stored in the result list to determine whether the value number S of k objects in the result list is greater than or equal to the comparison number V.
  • the selection device 4 at program point 37 outputs the k objects with the best value numbers via the formatting device 5 as a result to the input / output device 1. The program then ends.
  • the selection device 4 determines the missing properties by direct access to the database 3 and stores it in the result list.
  • the results of the direct accesses are shown in the access list of FIG. 11, which is stored in the data memory 6.
  • the selection device 4 then calculates at program point 39 the value number S (o) (aggregated score) for each object o and stores it m in the result list.
  • Figure 12 shows the value numbers of the result list. The program then branches to item 35.
  • the selection device 4 selects the object with the largest value from the data list with the lowest indicator, which is from this data list has not yet been selected (program item 33, program item 40), and stored in the result list. Then, at program point 41, the comparison device V is recalculated by the selection device 4, taking into account the newly selected object.
  • the following query at program point 42 checks whether the value number S (o) of k objects in the result list is greater than or equal to the comparison number V. If this is the case, the program branches to program item 37. If this is not the case, the following program item 43 recalculates the indicator for the data list from which the new object was selected at program item 40. The program then branches to program point 34.
  • the second algorithm has an increase in efficiency compared to the first algorithm. By evaluating the abort condition twice, fewer direct accesses are necessary. In addition, when selecting new objects which are included in the result list, the k best objects are determined very quickly by selecting the data list which has the largest indicator I. This effect stems from the fact that the probability that the comparison number V with an object from the data list with a large indicator quickly becomes smaller is greater than with an object from a data list with a small indicator.
  • FIGS. 9 and 10 show the two data lists that the search engine 2 determines from the database 3 and makes available to the selection device 4 at program point 32.
  • the objects ol, o2, o4 and o5 are selected by the selection device 4 and stored in the data memory 6 with the values (score).
  • lem ⁇ chtung 4 are sought via direct access to the database 3. The result of the direct access is shown in FIG. 11.
  • the selection device 4 calculates the respective value number S (aggregated score) of the objects according to program point 39 and stores them in the data memory 6 in accordance with a result list FIG. 12.
  • the termination condition according to program item 35 can be evaluated with the comparison number V, which is stored for each property m of the result list. Because the data lists are sorted, the lowest values show the objects that were last selected from the data lists: So here the objects o2 and o5: The comparison number is thus calculated as follows:
  • the query at program point 36 shows that the set of objects with a value number S (aggregated score)> comparison number V only consists of a single object ⁇ t, namely the object o4. So there is no termination.
  • the list of results at program point 40 must therefore be expanded. To do this, an object is fetched from the data list that has the larger indicator. In this case from the data list Si. The next object m of the data list s 1; Object o3 with a value S ⁇ (o3) of 0.85 that has not yet been read from this data list and is now being read out. The new minimum values of the two result lists thus provide the program track.
  • s 2 (o50 0.89 ,
  • the query at program point 42 shows that only the object o4 has a value number greater than or equal to the comparison number V. Thus the condition of the query at program point 42 is not fulfilled and the program branches to program point 43.
  • the program then branches to program item 37 and objects o4 and o5 are output as best objects in the entire database.
  • FIG. 13 shows a flow diagram of a third algorithm for determining k objects that correspond to a given object
  • sample object which is characterized by n properties, best of all.
  • n properties a combine function on F used to evaluate the properties for the comparison of the objects with the sample object.
  • the n properties and the combination function F are input via the input / output device 1 for the specified object.
  • the n properties are determined beforehand, for example, when the sample object is analyzed. Any combination function F can be used here.
  • the specified object, the specified properties and the combination function F correspond to those of the first algorithm according to FIG. 3.
  • the search engine 2 determines a data list from the database 3 at program point 51 for the properties texture and color, which are shown in FIGS. 14 and 15. The values of the properties of the objects are listed sorted by descending value. The data lists are fed to the selection device 4.
  • the selection device 4 selects from the supplied data lists a predetermined number m of values of each data list which represent the largest values of the data list and which have not yet been written into the result list.
  • the selected values are stored in the results list in the data memory 6 with the associated properties and identifications of the objects.
  • the selection device 4 compares the newly selected objects with each of the objects for which values have already been stored in the result list and decides which objects are identical. This check is particularly necessary in the case of heterogeneous information systems in which it is not clearly possible to assign the objects from the various data lists by means of the identification of the objects.
  • the comparison of the objects is carried out according to known methods, which are described, for example, by W. Cohen in the "Integration of Heterogeneous Databases without Common Domains Using Queries Based on Textual Similarity", Proceedings of ACM SIGMOD ⁇ 98, Seattle 1998.
  • the values of the properties are stored in the result list in the data memory 6 for all newly selected objects.
  • the values of properties that have not yet been recorded are estimated with the lowest value of the property that has occurred to date.
  • the value number is then calculated using the combination function F and entered in the access structure.
  • the selection device 4 checks whether k objects are completely known, i.e. whether k objects have actually determined and not estimated values for the properties for all properties to be considered. If this is not the case, the program branches back to program point 52.
  • program point 56 if the query at program point 56 reveals that k objects are already fully known in the properties under consideration, the program branches to program point 57.
  • program point 60 If the query at program point 58 shows that more than k objects are known, the program branches to program point 60.
  • the selection device 4 selects from all data lists a predetermined number of new objects which have the highest values for the data list (properties) and which have not yet been selected for this data list (property).
  • the values of the newly selected objects are assigned to an object in a manner analogous to program point 53 via a predefinable comparison function and m the result list is written into data memory 6. The values of the properties of the newly selected objects that cannot be assigned to an object already stored in the result list are discarded and no longer used.
  • the unknown values of the properties of the objects stored in the m result list are estimated in accordance with program point 55 with the known, minimum values of the properties, and m the result list is entered.
  • the value numbers S are calculated in accordance with program point 55 using the values newly entered in the result list.
  • program items 60, 61 and 57 no new objects m are entered in the results list, but only new values from objects already stored in the results list are fetched from the data lists and used for further estimation. The program then branches to program item 57.
  • the third algorithm according to FIG. 13 is explained in more detail below using an example:
  • the combination function F is the arithmetic mean of the texture and the color.
  • the specified object with the specified properties and the combination function corresponds to the specified object of the first algorithm.
  • the object ol and o2 with the respective greatest value of the texture or color m property the result list is entered.
  • the identifier and the value of the property are entered for each object.
  • the objects ol and o4 are processed in accordance with the program points 53, 54 and 55 and the value number S (aggregated score) is written into the access structure in accordance with FIG. 16.
  • the query at program point 56 shows that no k objects are completely known yet. Consequently, the other two objects o2, o5 at program point 52 are fetched from the data lists in FIGS. 14, 15 and the result list is entered with the identifier and the value of the property m. Processing program points 53, 54 and 55, the value number S is calculated for each object and written into the access structure according to FIG. 17.
  • the query at program point 56 again shows that k objects are not yet completely known.
  • the other objects o3, o6 are fetched from the data lists at program point 52 and the result list is entered with the identifier and the values for the properties m.
  • the value number S is calculated for the newly selected objects and m the access structure according to FIG. 18 is written in.
  • object o5 is next read from the first data list (FIG. 14) and object o3 from the second data list (FIG. 15) and the results list is entered with properties m.
  • the value numbers S for the object o5 and o3 are recalculated and written into the access structure according to FIG. 20.
  • the following query at program point 56 shows that three objects (o4, o5, o3) / complete are known in the result list.
  • the objects are removed in which the value number (aggregated score) was determined at least with an estimated value and the value number is smaller than the smallest value number of a fully known object. All objects except objects o4 and o5 are deleted from the results list.
  • the objects o4, o5 thus remain as the objects which are output as a result of the query after the processing of the program points 58 and 59.
  • An advantage of the third algorithm is that time-consuming direct access is avoided, particularly in the case of heterogeneous information systems. This results in a faster search algorithm.
  • the fourth algorithm essentially consists of two phases.
  • the first phase new objects are written into the result list and compared with the other objects.
  • the second phase can preferably be started after the appearance of the first k elements for all properties m of the result list.
  • this does not require time-consuming direct access to the objects in the database. Instead, the list of results for the properties only has to be expanded up to certain, geometrically estimated limit values with objects, the objects compared with one another and the value numbers calculated. to guarantee the correctness of the best objects.
  • the values (S x , ..., S) correspond to the values of the properties of the object of the results list, which has the smallest number of values and of which all values of the properties are known.
  • the values (Sj . , ..., S n ) correspond to the smallest values of the properties which are stored in the result list, ie the smallest known values of the properties.
  • the value C 0 corresponds to the value number (aggregated score) of the smallest object, the properties of which are all known and are stored in the result list.
  • the new entry in the results list is used with the previously used for the other properties in objects that have occurred in the result list, which essentially corresponds to a main memory operation of low complexity. If k objects already appear in the result list for all other properties, the second step, depending on the combination function F for all properties, is to load the objects from the data lists into the result list whose value numbers are greater than the value numbers of the previously calculated limit values S xl to S ⁇ are.
  • the objects newly entered in the results list are compared with the objects already stored. All objects that are known for all properties in the result list are sorted according to their value numbers and the first k objects can be output as a result of the search.
  • FIG. 22 shows a flowchart of the fourth algorithm, with which a predetermined number k of objects from a database is determined which best resemble a predetermined object (sample object).
  • n predefinable properties and a combination function F are entered via input / output device 1 for the specified object (sample object).
  • the predefined object, the predefinable properties and the combination function F correspond to those of the second algorithm according to FIG. 3.
  • the search engine 2 determines a data list for the texture and color properties from the database 3 at program point 71, which are shown in FIGS. 23 and 24.
  • the objects are sorted according to the descending value of the properties.
  • the data lists are fed to the selection device 4.
  • the selection device 4 selects from the supplied data lists a predeterminable number m of objects from each data list which have the largest values of the data list (properties) and whose values for this data list have not yet been entered into a result list in the data memory 6 were entered.
  • the values of the properties and the identifications of the objects are then stored in the results list in the data memory 6.
  • the selection device 4 compares the object identifiers newly entered in the result list with each of the object identifiers already stored in the result list and decides via a comparison function which object identifiers from different data lists belong to a single object.
  • the adjustment is carried out with the same function as in program point 53 of the third algorithm in FIG. 13.
  • a new access structure corresponding to FIG. 25 is created for each new object for which no values have yet been stored in the result list, in which the identifier of the object and the information about which property of the object is known are stored.
  • the selection device 4 writes all values of the properties of the new object that have been newly read at program point 73 into the access structure.
  • the selection device 4 then checks at program point 76 whether values are known for k objects in all properties to be considered. If this is not the case, the program branches back to program point 72. If the query at program point 76 shows that for k objects all values of the relevant properties in the result list are known, ie k objects are completely known, the program branches to program point 77. Instead of the number k, another number can also be used as a criterion in order to branch to program point 77.
  • the selection device 4 determines the value limits by forming a level hyper area in order to ensure that sufficient objects are considered so that a reliable statement about the best objects can be made. For this purpose, the selection device 77 selects the values of the object stored in the result list with the smallest number of values for determining the sufficient level hyper area. The system of equations described above with n equations for the combination function F is then solved for the values of the selected smallest object at program point 78.
  • the selection device 4 selects all objects of the data lists up to the associated value S X1 at program point 79 and the results list is written in the objects with values greater than the limit value S X1 .
  • the newly registered objects are compared in accordance with program point 73 with the objects previously seen and each object is uniquely assigned to an object.
  • the selection device 4 at program point 80 determines the k best fully known objects from the objects stored in the result list and outputs them at program point 81 via the formatting device 5 and the input / output device 1 as k best objects.
  • FIG. 24 a data list for the texture property and FIG. 24 a data list for the color property, which are ascertained by the search engine 2 and are transferred to the selection device 4.
  • Objects ol and o4 are read out from the data lists in FIGS. 23, 24 one after the other in accordance with program points 72 to 76 and a result list m is written into data memory 6.
  • the identifier of the object and the value of the property of the object are stored in the result list.
  • an access structure according to FIG. 25 is stored in the data memory 6.
  • An identifier for the object, the value number (aggregated score) of the object and the information about which property of the object is known are stored in the access structure.
  • the query at program point 76 shows that the program branches back to program point 72 and other objects are selected alternately from both data lists and processed according to program points 73, 74 and 75 and lr. the data memory 6 are written in until the values for the texture and color property are stored in the result list for k objects.
  • Figure 26 shows this status based on the access structure. From Fig. 26 it can be seen that the properties of the objects o4, o5 are completely known, so that after the program query at program point 76, the program branches to program point 77. The adequate level line can therefore be determined at program point 78. For this, as described above, the n equations for the combination function F must be solved
  • the object o7 cannot belong to the two best objects.
  • the next object from s- *. is the object o9 with a value of 0.75 and is therefore outside the limit of 0.77, which is calculated via the level hyper area was not. The object o9 is therefore no longer to be taken into account.
  • the methods according to the invention are preferably stored on a storage medium which can be read by a computer, so that the computer can process the methods.
  • a simple implementation of the device for carrying out the method consists of a computer which has the program blocks shown in FIG. 1, either hardware and / or software implemented.
  • the combination function F can be optimized depending on the sample object and the type of the given information systems in order to obtain the best possible search result.
  • the combination function allows the properties to be weighted, which can be entered individually.

Landscapes

  • Engineering & Computer Science (AREA)
  • Theoretical Computer Science (AREA)
  • Library & Information Science (AREA)
  • Databases & Information Systems (AREA)
  • Data Mining & Analysis (AREA)
  • Physics & Mathematics (AREA)
  • General Engineering & Computer Science (AREA)
  • General Physics & Mathematics (AREA)
  • Computational Linguistics (AREA)
  • Information Retrieval, Db Structures And Fs Structures Therefor (AREA)

Abstract

L'invention concerne des procédés permettant de trouver, de façon efficace, dans une pluralité d'objets, les objets qui ressemblent le plus à un objet modèle. A cet effet, le nombre des objets à examiner est limité à des valeurs limites calculées efficacement. En outre, ces procédés font appel à des stratégies de recherche qui exploitent les valeurs des caractéristiques des objets examinés pour être efficaces.
EP01911439A 2000-02-08 2001-02-08 Dispositif, support d'informations et procede pour trouver des objets presentant une grande similitude par rapport a un objet predetermine Withdrawn EP1254415A1 (fr)

Priority Applications (1)

Application Number Priority Date Filing Date Title
EP01911439A EP1254415A1 (fr) 2000-02-08 2001-02-08 Dispositif, support d'informations et procede pour trouver des objets presentant une grande similitude par rapport a un objet predetermine

Applications Claiming Priority (4)

Application Number Priority Date Filing Date Title
EP00102651A EP1124187A1 (fr) 2000-02-08 2000-02-08 Appareil, support d'enregistrement et méthode pour retrouver des objets ayant une forte similarité avec un objet donné
EP00102651 2000-02-08
EP01911439A EP1254415A1 (fr) 2000-02-08 2001-02-08 Dispositif, support d'informations et procede pour trouver des objets presentant une grande similitude par rapport a un objet predetermine
PCT/DE2001/000518 WO2001059609A1 (fr) 2000-02-08 2001-02-08 Dispositif, support d'informations et procede pour trouver des objets presentant une grande similitude par rapport a un objet predetermine

Publications (1)

Publication Number Publication Date
EP1254415A1 true EP1254415A1 (fr) 2002-11-06

Family

ID=8167802

Family Applications (2)

Application Number Title Priority Date Filing Date
EP00102651A Withdrawn EP1124187A1 (fr) 2000-02-08 2000-02-08 Appareil, support d'enregistrement et méthode pour retrouver des objets ayant une forte similarité avec un objet donné
EP01911439A Withdrawn EP1254415A1 (fr) 2000-02-08 2001-02-08 Dispositif, support d'informations et procede pour trouver des objets presentant une grande similitude par rapport a un objet predetermine

Family Applications Before (1)

Application Number Title Priority Date Filing Date
EP00102651A Withdrawn EP1124187A1 (fr) 2000-02-08 2000-02-08 Appareil, support d'enregistrement et méthode pour retrouver des objets ayant une forte similarité avec un objet donné

Country Status (5)

Country Link
US (1) US20030109940A1 (fr)
EP (2) EP1124187A1 (fr)
JP (1) JP2003527684A (fr)
AU (1) AU2001240461A1 (fr)
WO (1) WO2001059609A1 (fr)

Families Citing this family (6)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US7979451B2 (en) * 2008-03-19 2011-07-12 International Business Machines Corporation Data manipulation command method and system
US7979470B2 (en) * 2008-03-19 2011-07-12 International Business Machines Corporation Data manipulation process method and system
US8406573B2 (en) * 2008-12-22 2013-03-26 Microsoft Corporation Interactively ranking image search results using color layout relevance
US20110191334A1 (en) * 2010-02-04 2011-08-04 Microsoft Corporation Smart Interface for Color Layout Sensitive Image Search
US10467195B2 (en) 2016-09-06 2019-11-05 Samsung Electronics Co., Ltd. Adaptive caching replacement manager with dynamic updating granulates and partitions for shared flash-based storage system
US10455045B2 (en) 2016-09-06 2019-10-22 Samsung Electronics Co., Ltd. Automatic data replica manager in distributed caching and data processing systems

Family Cites Families (10)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US6182069B1 (en) * 1992-11-09 2001-01-30 International Business Machines Corporation Video query system and method
US5802361A (en) * 1994-09-30 1998-09-01 Apple Computer, Inc. Method and system for searching graphic images and videos
US5915250A (en) * 1996-03-29 1999-06-22 Virage, Inc. Threshold-based comparison
US6285788B1 (en) * 1997-06-13 2001-09-04 Sharp Laboratories Of America, Inc. Method for fast return of abstracted images from a digital image database
US6567551B2 (en) * 1998-04-27 2003-05-20 Canon Kabushiki Kaisha Image search apparatus and method, and computer readable memory
US6463432B1 (en) * 1998-08-03 2002-10-08 Minolta Co., Ltd. Apparatus for and method of retrieving images
JP2000148794A (ja) * 1998-08-31 2000-05-30 Canon Inc 画像検索装置及びその方法、コンピュ―タ可読メモリ
US6477269B1 (en) * 1999-04-20 2002-11-05 Microsoft Corporation Method and system for searching for images based on color and shape of a selected image
US6418430B1 (en) * 1999-06-10 2002-07-09 Oracle International Corporation System for efficient content-based retrieval of images
US6563959B1 (en) * 1999-07-30 2003-05-13 Pixlogic Llc Perceptual similarity image retrieval method

Non-Patent Citations (1)

* Cited by examiner, † Cited by third party
Title
See references of WO0159609A1 *

Also Published As

Publication number Publication date
EP1124187A1 (fr) 2001-08-16
WO2001059609A1 (fr) 2001-08-16
JP2003527684A (ja) 2003-09-16
AU2001240461A1 (en) 2001-08-20
US20030109940A1 (en) 2003-06-12

Similar Documents

Publication Publication Date Title
EP2187340B1 (fr) Dispositif et procédé pour déterminer un histogramme de côtés, dispositif et procédé de classement d'une image dans une base d'images de données, dispositif et procédé pour trouver deux images similaires et programme d'ordinateur
DE19513960A1 (de) Abbildung eines Graphen in einen Speicher
DE10034694A1 (de) Verfahren zum Vergleichen von Suchprofilen
EP1254415A1 (fr) Dispositif, support d'informations et procede pour trouver des objets presentant une grande similitude par rapport a un objet predetermine
DE2813157C2 (de) Gerät zur selbsttätigen, lageunabhängigen Mustererkennung
DE10017551A1 (de) Verfahren zur zyklischen, interaktiven Bildanalyse sowie Computersystem und Computerprogramm zur Ausführung des Verfahrens
AT522281B1 (de) Verfahren zur Charakterisierung des Betriebszustands eines Computersystems
WO2010078859A1 (fr) Procédé pour déterminer une similarité entre des documents
DE102020207449A1 (de) Verfahren, Computerprogramm und Vorrichtung zum Verarbeiten von Signalen
WO1997036248A1 (fr) Procede de determination de poids aptes a etre elimines, d'un reseau neuronal, au moyen d'un ordinateur
EP0978052B1 (fr) Selection assistee par ordinateur de donnees d'entrainement pour reseau neuronal
EP1273007A1 (fr) Procede permettant de determiner un enregistrement caracteristique pour un signal de donnees
DE102017118996B3 (de) Verfahren zur Bestimmung von einflussführenden Parameterkombinationen eines physikalischen Simulationsmodells
DE4238772C1 (de) Verfahren zur Auswertung einer Menge linguistischer Regeln
EP1111515A2 (fr) Système analytique d'informations
DE19624614C2 (de) Verfahren zum Entwurf oder zur Adaption eines Fuzzy-Reglers oder eines Systems von verknüpften Fuzzy-Reglern
DE60309191T2 (de) System zum fuzzy-assoziativen beschreiben von multimedia-gegenständen
DE19549300C1 (de) Verfahren zur rechnergestützten Ermittlung einer Bewertungsvariablen eines Bayesianischen Netzwerkgraphen
DE10206658A1 (de) Verfahren zum Überprüfen einer integrierten elektrischen Schaltung
DE4495111C2 (de) Verfahren zur Bestimmung einer Menge von charakteristischen Merkmalen im Rahmen einer Objekterkennung
DE102014214851A1 (de) Computerimplementiertes Verfahren und Computersystem zur Durchführung einer Ähnlichkeitsanalyse
DE102007051612A1 (de) Verfahren und Vorrichtung zum automatisierten Vergleichen zweier Sätze von Messwerten
DE10030712B4 (de) Verfahren zur Differenzierung von durch Suchmaschinen im Rahmen einer Suchanfrage ermittelten Referenzen auf Dokumente
DE112020001526T5 (de) Informationsverarbeitungsvorrichtung, informationsverarbeitungsverfahren, informationsverarbeitungsprogramm und informationsverarbeitungssystem
DE102020203689A1 (de) Vorrichtung und Verfahren zur Verarbeitung einer Messgröße

Legal Events

Date Code Title Description
PUAI Public reference made under article 153(3) epc to a published international application that has entered the european phase

Free format text: ORIGINAL CODE: 0009012

17P Request for examination filed

Effective date: 20020805

AK Designated contracting states

Kind code of ref document: A1

Designated state(s): AT BE CH CY DE DK ES FI FR GB GR IE IT LI LU MC NL PT SE TR

AX Request for extension of the european patent

Free format text: AL;LT;LV;MK;RO;SI

17Q First examination report despatched

Effective date: 20030128

STAA Information on the status of an ep patent application or granted ep patent

Free format text: STATUS: THE APPLICATION IS DEEMED TO BE WITHDRAWN

18D Application deemed to be withdrawn

Effective date: 20070901