JP7129206B2 - System, Aggregation Method and Program - Google Patents

Description

The present invention relates to a system, an aggregation method, and a program.

2. Description of the Related Art Conventionally, in printing systems, an output management system and a device management system having functions of totaling the number of prints and reporting have been widely introduced for the purpose of visualization and reduction of printing costs. In particular, understanding printing costs is one of the important propositions in cost management for medium-sized companies and offices that introduce office machines such as multiple printers and multifunction peripherals (MFPs). is.

For example, a reporting function is known that totals the number of printed sheets by a plurality of methods, such as by department, by user, and by application. By using such a reporting function, it is possible to determine whether or not the number of printed sheets for each department or each user is appropriate in light of the business content. In spite of similar business content, it is recommended that departments and users who print a large number of pages reduce their use of printing.
In addition, by examining the applications that print a large number of sheets, it is possible to infer in which field of work a lot of printing is performed, and to consider measures such as improving the work and promoting digitization.

On the other hand, there is known a technique of classifying electronic documents by applying machine learning techniques.
Patent Literature 1 discloses a technique for clustering text documents and labeling the clusters.
Further, Japanese Patent Application Laid-Open No. 2002-200002 discloses a technique of performing clustering using a plurality of features included in print data as feature amounts, and further finding clusters and print settings specific to the clusters.

In the field of machine learning, clustering and classification are two widely known techniques for classifying data such as electronic documents and print data.
Classification is a technique for learning by designating in advance what kind of group data is to be divided into. Usually, class classification takes a configuration in which a plurality of sets of sample data (teacher data) and their classification results (correct values) are prepared and learning is performed in advance. If the type of electronic document to be classified is specifically determined, such as by identifying a specific form format, classification can be performed with a high probability.
Clustering, on the other hand, is a technique for dividing a target data set into groups with similar characteristics. Clustering has the advantages of not requiring preparation of training data in advance, enabling flexible classification according to individual environments, and eliminating the need for troublesome operations such as prior learning. However, it is necessary to determine from the classified data what kind of meaning the clusters resulting from the classification have.

JP 2009-151390 A JP 2015-202667 A

With the conventional technology, even if the number of printed sheets for each application is known, it is difficult to specify the type of business that generates printing costs. For example, even among documents for the same electronic document viewer application, it is common for there to be presentation materials, forms, and documents with different contents. Also, it is conceivable that the same type of form is prepared for each of the spreadsheet application and the electronic viewer application.
In this way, the application and the type of business incurring printing costs are not necessarily linked. Therefore, it cannot be said that the conventional aggregation for each application is sufficient as information for reducing printing costs.

The present invention classifies the print data into one cluster out of a plurality of clusters based on the feature amount representing the feature of the print data using a multidimensional vector. and a tallying means for tallying up the total number of printed sheets of print data belonging to each of the plurality of clusters.

It is possible to classify the print data into clusters according to the content of the print data based on the feature amount that expresses the features of the print data with a multidimensional vector, and to grasp the total number of printed sheets for each cluster.

1 is a configuration diagram of a printing system; FIG. 2 is a hardware configuration diagram of a printer; FIG. 2 is a hardware configuration diagram of a server; FIG. 3 is a configuration diagram of software modules of the printer; FIG. 4 is a software module configuration diagram of the server; FIG. 4 is a basic flow chart of print data clustering in the server; 4 is a printing basic flow chart. 4 is a list of print settings based on job tickets; 4 is a detailed flowchart of print processing; 4 is a list of drawing commands by PDL; 4 is a detailed flowchart of feature extraction; It is a configuration example of a feature amount by a multidimensional vector. 4 is a detailed flowchart of clustering of printed data; It is an example of a tally report. FIG. 11 is a software module configuration diagram of the printer (Embodiment 2); 10 is a basic flowchart of printer processing (Embodiment 2); 4 is a detailed flow chart of text-based feature quantity extraction;

BEST MODE FOR CARRYING OUT THE INVENTION The best mode for carrying out the present invention will be described below with reference to the drawings.

<System configuration diagram>
FIG. 1 is a system configuration diagram showing a configuration example of a printing system according to this embodiment.
A printer A 100, a printer B 101, a printer C 102, a server 103, and a client PC 104 are connected to the network 105, and these devices are in a state of being able to communicate with each other.
The server 103 plays a role of an information processing device in this embodiment.
In this embodiment, three printers A 100, B 101, and C 103 are connected, but the number of printers is not limited to three, and more printers may be connected.
Also, instead of the printer, an image processing apparatus such as a multifunction machine having a print function may be connected. Similarly, a plurality of client PCs may also be connected.

<Printer hardware configuration diagram>
FIG. 2 is a hardware block diagram showing the hardware configuration of the printer A100 of this embodiment.
Here, only the printer A100 will be described, but the printers B101 and C102 also have the same hardware configuration as the printer A100.

The controller unit 200 controls input/output of image signals and device information.
The CPU 201 reads a program stored in the ROM 203 or HDD 204 to the RAM 202 and executes it. Furthermore, the CPU 201 centrally controls each device connected to the system bus 205 .
A RAM 202 is the main memory of the CPU 201 and is used as a work area for a control program that controls the printer.
The ROM 203 stores a boot program that is executed when the power is turned on.
The HDD 204 stores an operating system and a printer control program body. The HDD 204 is also used for temporarily or long-term storage of large-capacity data such as bitmap images and print data.

A network 206 connects to a local area network (LAN) 213 and is responsible for inputting/outputting print data and device information.
An operation unit I/F 207 is an interface unit with the operation unit 214 and outputs bitmap data to be displayed on the operation unit 214 to the operation unit 214 . It also plays a role of transmitting information input by the user of the printer A 100 from the operation unit 214 to the CPU 201 .
The operation unit 214 has a liquid crystal panel and a sound source as output devices, and a touch panel and hard keys as input devices.

Controller unit 200 is connected to printer engine 215 via device I/F 208 . A device I/F 208 transmits image signals, device operation instructions, and receives device information based on instructions from the CPU 201 .
The printer engine 215 is an output machine that outputs an image signal from the controller unit 200 onto a medium, and may be of either an electrophotographic method or an inkjet method.

A RIP (Raster Image Processor) 209 is dedicated hardware that develops a display list into a bitmap image. The RIP 209 processes the display list generated by the CPU 201 on the RAM 202 at high speed and in parallel with the execution of the CPU 201 .
A printer image processing unit 210 performs image correction, halftoning, and the like on print output image data.
The image companding/decompressing unit 211 performs JPEG compression/decompression processing for multilevel image data, and JBIG, MMR, and MH compression/decompression processing for binary image data.
The image rotation unit 212 rotates image data.

<Server hardware configuration diagram>
FIG. 3 is a hardware block diagram showing the hardware configuration of the server 103 of this embodiment.
The server 103 has a controller unit 300 , an operation section 310 and a display 311 .

Controller unit 300 includes CPU 301, RAM 302, ROM 303, HDD 304, network 306, operation unit I/F 307, display I/F 308, and the like.
The CPU 301 reads a program stored in the ROM 303 or HDD 304 to the RAM 302 and executes it. Further, the CPU 301 comprehensively controls each device connected to the system bus 305 .

A RAM 302 is the main memory of the CPU 301 and is used as a work area for various programs.
A ROM 303 stores a boot program that is executed when the power is turned on.
HDD 304 stores an operating system and application programs. The HDD 304 is also used for the purpose of temporarily or long-term holding large amounts of data.

An operation unit I/F 307 is an interface with an operation unit 310 including an input device such as a mouse and a keyboard, and notifies the CPU 301 of information input through the operation unit 310 .
The display I/F 308 outputs image data to be displayed on the display 311 to the display 311 .
A network 306 connects to a local area network (LAN) 309 and is responsible for communication with external devices such as client PCs and printers.

<Printer software module configuration diagram>
FIG. 4 is a software module configuration diagram for explaining the software configuration of the printer A100 of this embodiment.
Here, only the printer A100 will be described, but the printers B101 and C102 also have the same software configuration as the printer A100.
Each software module shown in FIG. 4 is stored in HDD 204 as a program, loaded into RAM 202 and executed by CPU 201 . More specifically, each software module is loaded into the RAM 202 by an OS (Operating System) running on the CPU 201, given execution rights in units of threads, and executed.

A data receiving unit 402 receives print data transmitted from the server 103 . The received data is held by the job data management unit 409 via the job control unit 401 .
A job control unit 401 manages overall job control from data reception to printing.
A PDL interpreter 403 interprets PDL data described in a page description language (PDL) as print data and generates a display list as intermediate data. The generated display list is held by the job data management unit 409 via the job control unit 401 .
A renderer 404 is a module that generates a bitmap image from a display list. Much of the processing is performed by dedicated hardware RIP 209 . The generated bitmap image is held by the job data management unit 409 via the job control unit 401 .

A print driver 406 sends a print instruction and a bitmap image to the printer engine via the device I/F 208 . At this time, the print driver 406 also performs image correction by the printer image processing 210 .
A user interface 405 is a module that controls the operation unit 214 via the operation unit I/F 207 . It also generates data to be displayed on the liquid crystal panel of the operation unit 214 and updates the display of the liquid crystal panel according to the input from the touch panel. Also, if the input from the touch panel is an instruction to execute a certain job, the instruction is transmitted to the job control unit 202 .

A feature amount extraction unit 407 is a module that analyzes PDL data and extracts feature amounts. The extracted feature amount is held by the job data management unit 409 via the job control unit 401 . Details of the feature amount will be described later.
A data transmission unit 408 is a module that transmits feature amount data held in the job data management unit 409 to the server after printing is completed.
A job data management unit 409 is a database that retains and manages print data, display lists, bitmap images, and feature amount data, either temporarily or over the long term.

<Server software module configuration diagram>
FIG. 5 is a software module configuration diagram showing the software configuration of the server 103 of this embodiment.
Each software module shown in FIG. 5 is stored in the HDD 304 as a program, loaded into the RAM 302 and executed by the CPU 301 . More specifically, each software module is loaded into the RAM 302 by an OS (Operating System) running on the CPU 301, given execution rights in units of threads, and executed.

A feature amount data receiving unit 505 receives feature amount data transmitted from the printer. The received data is held in data management section 506 via control section 501 .
The control unit 501 manages a series of processes from reception of feature amount data to clustering and report generation.
A clustering execution unit 502 is a module that executes data clustering of print data using feature amount data.
A report generation unit 503 is a module that generates a total report of clusters of print data and the number of prints. A request for report generation is received from the Web server 504 . The generated report is sent back to the requester by web server 504 .

<Basic flow chart for print data clustering on the server>
FIG. 6 is a basic flowchart for clustering print data in the information processing apparatus (server 103).
It should be noted that the clustering will be explained here assuming that it is executed by each software module in FIG. 5 stored in the HDD 304 is read to the RAM 302 and executed by the CPU 301, thereby realizing this flowchart.

First, in S601, the control unit 501 waits until a server event occurs.
There are two types of server events: a notification of reception of feature amount data, which will be described later, and a request for clustering of print data and generation of a tally report of the number of prints.

When a server event occurs, the process advances to step S602, and the control unit 501 determines whether the server event is notification of reception of feature amount data.
In the case of Yes in S<b>602 , the process proceeds to S<b>603 , and the feature amount data reception unit 505 receives the feature amount data and stores the data in the data management unit 506 via the control unit 501 .
A feature amount is a feature of print data represented by a multidimensional vector, and is extracted during preprocessing for printing. In addition to the feature amount, the feature amount data includes, as attached information, the number of prints, the number of color prints, the number of monochrome prints, and a representative image of the print data that is the source of the feature amount. Further details of the feature amount data and its extraction method will be described later.

After completing the process of S603, the process returns to S601, and the control unit 501 again waits for a server event.
If No in S602, the process advances to S604, and the control unit 501 determines whether the server event is a request for clustering of print data and generation of a tally report of the number of prints.
It should be noted that a request for generation of a tally report generated by a user's operation on a Web page by the Web server 504 is notified to the control unit 501 as an event.

If Yes in S604, the process advances to S605, and the clustering execution unit 502 executes data clustering of the print data using the feature amount in order to classify the printed data into a plurality of clusters.
The target feature amount data is all the feature amount data saved in the data management unit 506, but filtering may be performed according to the period during which the feature amount data was received, the printer from which the feature amount data was sent, or the like.
Data clustering is a process of dividing data into clusters, which are groups of data with similar feature values, and the detailed flow will be described later with reference to FIG. 13 .
As a result of data clustering, a plurality of clusters and each feature amount data, that is, the cluster to which the original print data belongs is determined. A cluster and a list of feature amount data belonging to the cluster are stored in the data management unit 506 via the control unit 501 .

Next, in S606, the report generation unit 503 counts the number of printed sheets for each cluster.
The report generation unit 503 calculates the total number of printed sheets of print data belonging to each cluster, the total value of color printing, and the total value of monochrome printing.

Next, in step S607, the report generation unit 503 generates an electronic document including a clustering report of the print data and a tabulation report of the number of printed sheets.
The report generation unit 503 writes the total number of pages printed, the total value of color printing, and the total value of monochrome printing of each cluster in a total report. The report generation unit 503 also displays a representative image, which is print data closest to the center of gravity of the cluster, as an image that best reflects the characteristics of each cluster.
FIG. 14 shows an example of a total report. In this example, four clusters are assigned to categories 1 to 4, respectively, and a representative image and the number of prints are displayed for each.

Next, in step S608, the Web server 504 sends back to the requesting user a print data clustering report and a tally report of the number of prints.
Then, the process returns to S601, and the control unit 501 waits for a server event again.
If No in S604, the control unit 501 similarly returns to S601 and waits for a server event again.

Although the data clustering according to this flowchart is executed at the time of a user request via a web page, it may be executed at night or during an idle time when there is no user request.

<Print basic flow chart>
FIG. 7 is a basic flow chart of printing in the printer.
It should be noted that, here, it will be described as being executed by each software module in FIG. 4 stored in the HDD 204 is read to the RAM 202 and executed by the CPU 201, thereby realizing this flowchart.

First, in step S<b>701 , the data reception unit 402 receives print data and stores the print data in the job data management unit 409 via the job control unit 401 .
Print data is generated by an application and printer driver on the host PC and sent to the printer. The print data consists of PDL (page description language) data and a job ticket.
PDL expresses the content to be drawn on paper. The job ticket expresses print setting information such as the number of copies, double-sided printing, color/monochrome printing, and the like.

Next, in S702, the PDL interpreter 403 interprets the print data. Also, the feature amount extraction unit 407 performs feature amount extraction. A renderer 404 performs RIP processing. The renderer 404 also generates a thumbnail image of the first page of print data as a representative image.
The extracted feature amount and representative image are saved in the job data management unit 409 . Interpretation of print data, feature extraction, and RIP processing are executed in parallel, and the details will be described later with reference to FIG.

Next, in S703, the job control unit 401 executes printing of the print data.
In step S<b>704 , the job control unit 401 confirms the number of prints in the print data and saves it in the job data management unit 409 .
Note that the number of prints in the print data is linked to the feature amount of the print data and managed by the job data management unit 409 . Also, the number of printed sheets is determined by referring to a value managed by a printed number counter, which will be described later.
Then, in step S705, the data transmission unit 408 transmits feature amount data such as the feature amount of the print data, the number of prints, and the representative image to the server.

<List of print settings by job ticket>
FIG. 8 is a list of print settings that can be designated by a job ticket, which constitute print data together with PDL.
There are five print settings: "double-sided", "staple", "color mode", "paper size", and "page combination".

"Duplex" is a function that prints consecutive pages only on the front side of the paper, or alternately prints on the front side, the back side, the front side, and the back side, and can specify 1: 1 side and 2: 2 sides as an attribute value. is.
"Staple" is a function for stapling a plurality of output sheets, and as attribute values, 1: none, 2: single, and 3: double can be specified.
"Color mode" is a function of outputting print data in color or converting it to monochrome before outputting, and as attribute values, 1: automatic, 2: color, and 3: monochrome can be specified. Here, the attribute value 1: automatic is for the PDL interpreter 203 to automatically determine color or monochrome depending on the contents of the print data.

"Paper size" is a setting for specifying the output paper size, and 1: A4, 2: A3, and 3: B4 can be specified as attribute values.
"Combination of pages" is a function of reducing a plurality of consecutive pages in print data onto a designated sheet, laying them out, and printing them. As attribute values, 1:1in1, 2:2in1, 3:4in1, and 4:8in1 can be specified. For example, in the case of 2in1, two consecutive pages are laid out on one side of one output sheet.

<Detailed Flowchart of Print Processing>
FIG. 9 is a detailed flowchart of the printing process, and details the processing of S702 and S703 of the basic printing flowchart of FIG.
In this flowchart, three threads, thread A, thread B, and thread C, are executed in parallel (multithreading).
Each thread is time-divided and assigned its execution rights by the operating system. Since the time-sharing unit is sufficiently small, it can be considered that the three threads are operating in parallel.
Multi-threading processing by an operating system is a widely known technique, so detailed description thereof will be omitted. Thread A, thread B, and thread C are resident threads generated by the operating system when the print data classification apparatus is started.

In thread A, first, in step S901, the job control unit 401 resets the number-of-printed counter to 0, and extracts a job ticket describing print settings from print data.
Next, in step S<b>902 , the job control unit 401 determines whether the print data is print data of a job for which page-aggregate printing is instructed.
Here, page aggregate printing refers to a printing method such as N-UP printing in which a plurality of pages are arranged on a single sheet.

If Yes in S902, the process proceeds to S904.
Then, in S904, the PDL interpreter 403 calculates logical pages to be laid out on a single sheet and layout locations.
Next, in step S905, the PDL interpreter 403 processes the PDL data of the page assigned to the sheet according to the color mode setting and generates a sheet display list.
At this time, the PDL interpreter 403 adjusts the drawing position within the sheet of each page according to the calculated placement location. Note that the display list is intermediate data representing the drawing information of the sheet.
Then, the process proceeds to S906.

If No in S902, the process proceeds to S903.
In step S903, the PDL interpreter 403 processes the PDL data according to the color mode setting and generates a display list for one page.
Then, the process proceeds to S906.

In S906, the feature amount extraction unit 407 processes the PDL data and extracts feature amounts.
Details of the feature amount extraction processing will be described later with reference to FIG. 11 (and FIG. 17).

Next, in step S907, the PDL interpreter 403 notifies the renderer 404 of completion of generation of the display list and requests rendering.
In step S908, the PDL interpreter 403 determines whether or not all pages have been processed.
If Yes, the processing of this flow ends.
If No, return to S902 to continue processing the remaining pages.

Thread B's flow is executed by renderer 404 .
First, in S909, the renderer 404 waits for completion of generation of a display list for one sheet.
When the display list generation completion is notified in S907 of thread A, the process of thread B advances from S909 to S910.

At S910, renderer 404 renders to generate a bitmap image.
Here, the generated bitmap image has 8-bit gradation for each color of CMYK.
Also, the renderer 404 generates a thumbnail image in which the resolution of only the first image is lowered. Whether or not it is the first sheet is determined by the above-described printed number counter.

In step S<b>911 , the renderer 404 saves the bitmap image in the job data management unit 409 and requests the print driver 406 to print it.
A print request is executed by sending a rendering end notification to the print driver 406 . Since the print driver 406 executes processing in synchronization with the engine, the print request is executed as thread C, which is a thread separate from thread A.

The print driver 406 that has received the request in S911 of thread B transmits a print start request command to the engine and transfers the bitmap image in S913 of thread C. FIG.
At the same time, the print driver 406 transmits operation instruction commands for output paper size, duplex, and stapling. Prior to bitmap image transfer, the bitmap image is subjected to image processing.
Next, in S914, the print driver 406 increments the printed number counter except when outputting the back side of both sides.

When the process of S911 of thread B ends, the process advances to S912, and the renderer 404 determines whether rendering of all pages has been completed.
If Yes, the flow of FIG. 9 ends.
If No, return to S909 and repeat the process.
It should be noted here that the logical number of pages of print data does not match the number of output sheets. In N-UP printing, the number of output sheets is smaller than the number of logical pages. The same applies to double-sided printing.

<List of drawing commands by PDL>
FIG. 10 is a list of PDL drawing commands for describing print data in the print data classification device.
A drawing command is composed of DrawPath, DrawFillPath, DrawText, and DrawImage.

DrawPath is a command for drawing a line segment connecting a sequence of points configured by a coordinate array, and the line color and line width are specified as additional parameters. A pixel coordinate system with the upper left corner as the origin is used for the coordinates.
DrawFillPath is similarly a command for filling an area enclosed by a sequence of points configured by a coordinate array, and specifies the fill color as an additional parameter.
DrawPath and DrawFillPath can also draw curves by finely connecting line segments by point sequences.

DrawText is a command for drawing a character string with a specified font, and (character) size, color, and drawing position are specified as additional parameters.
DrawImage is a command for drawing a string of pixel data in bitmap format, and the size (width, height) and drawing position are specified as additional parameters.

<Detailed flow chart of feature extraction>
FIG. 11 is a detailed flow chart of feature quantity extraction, showing in detail the process of S906 in FIG.

In S1101, the feature amount extraction unit 407 parses (syntax analysis) the PDL data to extract a drawing command.
Next, in step S1102, the feature amount extraction unit 407 obtains approximate coordinates by rounding off the drawing coordinate values according to the drawing command.
Even if a similar template is used, the position may change to such an extent that it cannot be discerned from the appearance due to modification or inadvertent change of the template. For this reason, rounding processing is performed by rounding off in order to provide resistance to changes in coordinate values.

In S1103, the feature amount extraction unit 407 determines whether the drawing command is DrawPath or DrawFillPath.
If Yes, the process proceeds to S1105, and if No, the process proceeds to S1104.

If the drawing command is either DrawPath or DrawFillPath, in S1105, the feature quantity extraction unit 407 combines the drawing command, the approximate coordinates, and the number of points and extracts them as a drawing identifier.
Drawing identifiers classify drawings by DrawPath or DrawFillPath by position and the number of point sequences and handle them as features. That is, different drawing identifiers are treated as different features.
Since the drawing identifier is defined by only the approximate coordinates and the number of points, the same drawing identifier may be assigned to different drawings. However, since the probability is very limited, it is sufficiently possible to identify each document and each page by using a set of drawing identifiers as features.
Note that, in this embodiment, the feature amount is treated as a set of features.

If the drawing command is neither DrawPath nor DrawFillPath, in S1104 the feature amount extraction unit 407 determines whether the drawing command is DrawText.
If Yes, the process proceeds to S1106, and if No, the process proceeds to S1107.
If the drawing command is DrawText, in S1106 the feature amount extraction unit 407 combines the drawing command, approximate coordinates, and character string length to obtain a drawing identifier.

If the drawing command is not DrawText, in S1107 the feature amount extraction unit 407 combines the drawing command, approximate coordinates, width, and height to obtain a drawing identifier.
Note that S1107 is executed only when the drawing command is neither DrawPath, DrawFillPath, nor DrawText, that is, when it is DrawImage.

Proceeding to S1108, the feature amount extraction unit 407 increments and stores the count of the same drawing identifier.
Count is used as a value that represents the strength of a feature. The existence of a plurality of identical drawing identifiers in PDL data often means that the same drawing is performed on a plurality of pages. Template-based documents often use similar headers and footers on each page, so it can be said that the count of drawing identifiers takes into consideration the characteristics of the template. As another method, it is also possible to perform weighting according to the size of the drawing area.

Next, in step S1109, the feature amount extraction unit 407 determines whether or not the parsing of the PDL data has been completed.
If Yes, proceed to S1110.
If No, the process returns to S1101 to repeat the feature quantity extraction process.

When the parsing of the PDL data is completed, in S1110, the feature amount extraction unit 407 generates a multidimensional vector in which one drawing identifier is one dimension and the count of each drawing identifier is the value of each dimension.
This multidimensional vector becomes the feature quantity of the print data.

<Configuration example of feature amount by multidimensional vector>
FIG. 12 is a configuration example of a feature amount by a multidimensional vector of certain print data.
For example, a drawing identifier "DrawPath_100_80_2" with a dimension number of 1 means that there are two path drawings with a point sequence number of 2 at approximate coordinates (100, 80) by the DrawPath command.
The drawing identifier "DrawText_230_250_8" with dimension number 4 means that there are four text drawings with a character string length of 8 at approximate coordinates (230, 250) by the DrawText command.
The drawing identifier "DrawImage_850_1200_2400_1200" of dimension number 5 means that there is one drawing of an image with width: 2400 and height: 1200 at approximate coordinates (850, 1200) by the DrawImage command.

In the entire print data, there are 135 drawings identified by drawing identifiers, which are composed of 135-dimensional feature amounts.
In FIG. 12, a character string is used as the representation format of the drawing identifier, but any representation format may be used.

<Detailed flowchart for clustering of printed data>
FIG. 13 is a detailed flowchart of clustering processing for printed data, and shows the processing of S605 in FIG. 6 in detail.

In S1301, the clustering execution unit 502 determines a feature space for each piece of printed data from its feature amount.
Each printed data may or may not have a common drawing identifier. In order to compare feature quantities required for clustering, it is necessary to perform in a common feature space.

Therefore, first, a feature space that includes feature quantity identifiers of all print data is synthesized. Since each piece of printed data often has a different feature quantity identifier, this process greatly increases the dimensionality of the feature space relative to the number of dimensions of each piece of printed data. Therefore, a feature space with reduced dimensions is generated by deleting the feature quantity identifiers that exist only in specific print data. This is because a feature quantity identifier that exists only in specific printed data does not serve as an identifier representing a group feature.

Next, in step S1302, the clustering execution unit 502 maps the feature amount of each piece of printed data to feature points in the feature space.
Here, the clustering execution unit 502 deletes feature quantity identifiers that exist only in specific printed data. Printed data with an extremely large number of deleted feature quantity identifiers may be excluded from clustering. This is because an extremely large number of deleted feature quantity identifiers means that there is almost no rendering similar to other print data.

Next, in S1303, the clustering execution unit 502 determines the number of clusters and the representative point of each cluster by canopy clustering.
As will be described later, this embodiment employs the known K-means method as a clustering method.
In the K-means method, it is necessary to determine the number of clusters and the center of gravity of the initial clusters in advance. Canopy clustering is adopted as a preprocessing to properly determine the number of clusters and the centroids of the initial clusters. Since the canopy clustering is also a known technique, the explanation is omitted.

Next, in S1304, the clustering execution unit 502 sets the representative point of each cluster as the initial centroid point.
The clustering execution unit 502 executes clustering by the K-means method by the processing from S1305 to S1309.

In S1305, the clustering execution unit 502 calculates the distance between the feature point of each printed data and the center of gravity of each cluster.
Euclidean distance, which is a known method, is used as a distance calculation method. Here, it should be noted that the degree of distance represents the degree of similarity between feature quantities. That is, the distance between similar feature amounts (feature points) is small, and the distance between dissimilar feature amounts (feature points) is large.
Next, in step S1306, the clustering execution unit 502 assigns each piece of printed data to the cluster having the center of gravity closest to its feature point.
In S1307, the clustering execution unit 502 determines whether or not there is a change in assignment of feature points to clusters.
If Yes, go to S1308; if No, go to S1309.

If there is a change in the assignment of feature points to clusters, the clustering execution unit 502 recalculates the center of gravity of each cluster in S1308.
The centroid point is the average value of the feature points assigned to each cluster, and can be regarded as the feature point representing the feature of the cluster.
After completing the process of S1308, the process returns to S1305 to continue the process. These processes are repeated until there is no change in the assignment of feature points to clusters through the processes of S1308, S1305, and S1306, that is, until the assignment converges.

After the center of gravity of each cluster is determined, in S1309, the clustering execution unit 502 stores the feature amount of the center of gravity of each cluster and the printed data belonging to each cluster.
Then, in S1310, the clustering execution unit 502 checks and stores the printed data closest to the center of gravity of each cluster.
As the representative image of the cluster in S607 of FIG. 6, the image of the printed data closest to the center of gravity of the cluster is used.

Here, the printed data belonging to each cluster is the printed data assigned to any cluster having the center of gravity point closest to the feature point. In the K-means method, all feature points (feature amounts) are assigned to one of the clusters. Therefore, even an isolated point far from the center of gravity of any cluster is assigned to one of the clusters.
Since such isolated points become noise, in order to remove such noise, a method is adopted in which only feature points (feature amounts) within a certain distance from the centroid of the cluster are treated as printed data belonging to the cluster. You may In this case, the printed data that do not belong to any cluster can be collectively handled as other data, thereby avoiding the omission of information.

Since clustering evaluates similarity using distance in feature space, print data having feature points (feature amounts) having many identical features are extracted into the same cluster. That is, print data having many matching features can be identified as print data belonging to the same group from a large number of features included in the print data.

As a data clustering method, other methods such as hierarchical clustering and support vector machine may be used in addition to the K-means method in this embodiment. Even in that case, the similarity is evaluated by the distance between feature amounts.

As described above, according to this embodiment, it is possible to grasp the printing cost based on the content of the print data rather than the application that generated the print data.
Moreover, pattern recognition generally requires preparation in advance, and various patterns that occur after the fact cannot be taken into consideration afterward. For this reason, it is necessary to predetermine and program the pattern to be recognized as a specific format. That is, when a new pattern appears, programming is required again.

On the other hand, according to the present embodiment, a plurality of features included in print data are used as feature amounts to perform data clustering, thereby eliminating the need for programming for recognizing specific patterns in advance.
In addition, since learning using teacher data is not required, the method can be said to be extremely convenient. Furthermore, in order to determine the similarity of feature amounts consisting of a plurality of features, it is possible to recognize various formats without depending on specific features. That is, the clustering of this embodiment can be said to be a highly flexible classification method for various input print data.

In the first embodiment, a system that performs distributed processing by a server and clients (printers) has been described.
On the other hand, in the second embodiment, a system that performs clustering and report generation with a single printer will be described.

Also, in the first embodiment, a method of configuring the feature amount of print data using the drawing command type, position information, size information, and the like has been described. Such a feature amount is suitable for extracting data having a high degree of similarity in the arrangement and size of drawing objects. That is, it can be said that the ability to distinguish print data using similar template formats is high.
On the other hand, in a second embodiment, a method of using the content of text (printed character string) as the feature amount of print data will be described. By using text, it is possible to extract groups whose content of print data is semantically similar.
Here, it should be noted that there is no dependency between the server/client system configuration and the feature extraction method. That is, the feature amount extraction method of the first embodiment can be adopted even in the configuration of a single printer.

<Printer software module configuration diagram>
FIG. 15 is a software module configuration diagram showing the software configuration of the printer according to the second embodiment.
In the second embodiment, since the printer also functions as a server, some modules are added to the software module configuration diagram of FIG. Specifically, a clustering execution unit 410, a report generation unit 411, and a web server 412 are added.
The functionality of these modules is equivalent to the server clustering execution unit 502, report generation unit 503, and web server 504 in FIG. 5, respectively.

However, in the second embodiment, the data transmission unit 408 is not required in the printer because data such as feature amounts are not required to be transmitted/received between the server and the printer.
Further, since the processing capability of a printer alone is inferior to that of a server, in the case of a printer alone, the configuration is such that the total number of printed sheets is limited to the printer itself. That is, printing by other printers on the network is not counted. Others are the same as those of the software module configuration diagram of FIG. 4, so description thereof will be omitted.

<Basic flowchart of printer processing>
FIG. 16 is a flow chart showing basic processing in the printer according to the second embodiment.
First, in S1601, the job control unit 401 waits until a printer event occurs.
There are two types of printer events: a print data reception notification and a print data clustering and print count total report generation request.

When a printer event occurs, the process advances to step S1602, and the job control unit 401 determines whether the printer event is print data reception notification.
If Yes in S1602, the process advances to S1603 to receive print data.
If No in S1602, the process proceeds to S1608.

If the printer event is print data reception notification, the processing from S1603 to S1606 is the same as the processing from S701 to S704 in FIG. 7, so description thereof will be omitted.
In step S<b>1607 , the job control unit 401 saves the print data feature amount, the number of prints, and the representative image in the job data management unit 409 .
When the process of S1607 is completed, the process returns to S1601.

If No in S1602, that is, if the printer event is a request for clustering of print data and generation of a tally report of the number of prints, the process advances to S1608.
In step S1608, the clustering execution unit 502 clusters the printed data.

Since the processing from S1609 to S1612 is the same as the processing from S605 to S608 in FIG. 6, description thereof will be omitted.
When the process of S608 is completed, the process returns to S1601.

<Detailed Flowchart of Characteristic Value Extraction from Text>
FIG. 17 is a detailed flowchart of text-based feature quantity extraction, and shows in detail the processing of S906 in FIG. 9, similar to FIG.

In S1701, the feature amount extraction unit 407 parses (syntax analysis) the PDL data to extract a drawing command.
Next, in S1702, the feature quantity extraction unit 407 determines whether the drawing command is DrawText.
If Yes, the process proceeds to S1703, and if No, the process proceeds to S1706.
In S1703, the feature amount extraction unit 407 extracts character string information from the drawing command.
Note that, as described with reference to FIG. 4, DrawText has a drawing character string as a parameter.

In S1704, the feature amount extraction unit 407 morphologically analyzes the character string information and extracts only nouns as words.
Morphological analysis refers to a process of dividing a document into words and determining parts of speech using a dictionary. Here, a technique known as morphological analysis is used.
Next, in S1705, the feature amount extraction unit 407 stores the words and the cumulative number of occurrences of each word as features.

Next, in step S1706, the feature amount extraction unit 407 determines whether or not the parsing of the PDL data has been completed.
If Yes, the process advances to S1707.
If No, the process returns to S1701 to repeat the feature quantity extraction process.
When the parsing of the PDL data is completed, in S1707, the feature amount extraction unit 407 generates a multidimensional vector in which one word is one dimension and the number of occurrences of each word is the value of each dimension.

The same method as in the first embodiment may be adopted for processing other than the feature amount extraction. However, in Example 2, a more desirable result can be obtained by adopting the cosine distance than the Euclidean distance as the distance between feature quantities.
A more desirable result can be obtained by adjusting the weights of features by a known TF/IDF method as preprocessing for clustering (S1301 and S1302 in FIG. 9).

As described above, when text (printed character strings) is used as the feature quantity of print data, print data that use many of the same words are extracted as the same cluster. In other words, even if the formats are different, documents with similar semantic contents of the original documents of the print data are collected. For this reason, not only geometric characteristics such as drawing object placement, but also semantic characteristics can be focused on and examined for correlation with print settings to enable more multifaceted analysis.

In addition, since geometric properties and semantic properties can be regarded as independent, they can be treated simultaneously. In this case, some print data belongs to a cluster based on geometric characteristics and a cluster based on semantic characteristics at the same time.
In the first embodiment, the representative image of the cluster is displayed in the report, but in the second embodiment, a method of displaying a word characteristic of the cluster as meta information representing the characteristics of the cluster is also conceivable.

(Other examples)
The present invention supplies a program that implements one or more functions of the above-described embodiments to a system or apparatus via a network or a storage medium, and one or more processors in the computer of the system or apparatus reads and executes the program. It can also be realized by processing to It can also be implemented by a circuit (for example, ASIC) that implements one or more functions.
Moreover, the present invention may be applied to a system composed of a plurality of devices or to an apparatus composed of a single device.
The present invention is not limited to the above-described embodiments, and various modifications (including organic combinations of each embodiment) are possible based on the spirit of the present invention, and they are excluded from the scope of the present invention. not a thing In other words, the present invention includes all configurations obtained by combining each of the above-described embodiments and modifications thereof.

100 Printer A
103 Server 401 Job control unit 402 Data reception unit 403 PDL interpreter 404 Renderer 405 User interface 406 Print driver 407 Feature extraction unit 408 Data transmission unit 409 Job data management unit 501 Control unit 502 Clustering execution unit 503 Report generation unit 504 Web server 505 Feature amount data receiving unit 506 Data management unit

Claims (14)

a system,
an extraction means for extracting a feature amount of print data;
Classifying means for classifying the print data into one cluster out of a plurality of clusters based on a feature amount representing the feature of the print data as a multidimensional vector ;
and totaling means for totalizing the total number of printed sheets of print data belonging to each of the plurality of clusters. 2. The system of claim 1, wherein the print data comprises PDL data described in a page description language (PDL) and a job ticket. 3. The system according to claim 1, wherein the feature amount is represented by a multi-dimensional vector whose dimensions are each of the feature amount identifiers. 4. The system of claim 3, wherein the feature identifier is a drawing identifier derived from print data. 5. The system of claim 4, wherein the drawing identifier is defined from arithmetic coordinates and point sequences of print data described in a page description language. 4. The system of claim 3, wherein the feature identifiers are words extracted from print data. 7. The system according to any one of claims 3 to 6, wherein from among the feature quantity identifiers, a feature quantity identifier that exists only in specific print data is deleted. The classification means are
8. The system according to any one of claims 3 to 7, wherein the print data is classified into clusters having a centroid point closest to the feature amount. 9. The system according to any one of claims 1 to 8, further comprising generating means for generating a report of the number of printed sheets counted by the counting means. 10. The system according to claim 9, wherein said generating means generates a representative image from print data classified into each cluster. 11. The system according to claim 10, wherein the representative image is the print data closest to the center of gravity of the cluster among the print data classified into each cluster. 12. A system according to any preceding claim, further comprising printing means for printing based on said print data. An aggregation method comprising:
an extraction step of extracting a feature amount of print data;
a classification step of classifying the print data into one cluster out of a plurality of clusters based on a feature quantity representing the characteristics of the print data as a multidimensional vector ;
and a counting step of counting the total number of printed sheets of print data belonging to each of the plurality of clusters. A program for causing a computer to execute the counting method according to claim 13.

Images