Recording medium discrimination device and recording medium type discrimination method

Abstract

To reduce a frequency of imaging required for discrimination of a recording medium.SOLUTION: A recording medium type discrimination method discriminates a group of a recording medium based on a feature amount extracted from imaged image data using a first parameter, and discriminates a recording medium type based on the feature amount extracted from the image data using a second parameter.SELECTED DRAWING: Figure 7

Description

[Technical Field]

[0001] The present invention relates to a recording medium discrimination device and a recording medium type discrimination method. [Background technology]

[0002] There are known recording devices that can record images on various recording media such as glossy paper, semi-glossy paper, matte paper, etc. Some of these recording devices have parameters set according to the characteristics of the recording media in order to perform optimal recording for each type of recording media.

[0003] When a user sets the type of recording medium, the wrong type of recording medium may be set. If a recording medium with significantly different characteristics is set incorrectly, there is a risk that the amount of ink ejected may change, affecting the image.

[0004] Patent document 1 describes a method of determining a group of recording media using images of the recording media captured under first imaging conditions, and then using images captured under second imaging conditions based on the determined group, determining the type of the captured recording media from within the group. [Prior art documents] [Patent documents]

[0005] [Patent Document 1] Japanese Patent Application Laid-Open No. 2006-292952 Summary of the Invention [Problem to be solved by the invention]

[0006] However, since Patent Document 1 requires capturing images twice, it takes time to make the determination.

[0007] The present invention has been made in view of the above-mentioned problems, and has as its object to reduce the time required to identify the recording medium. [Means for solving the problem]

[0008] The present invention includes an image data acquisition means for acquiring image data obtained by imaging a predetermined area of ​​a recording medium, a first extraction means for processing the image data using a first parameter to extract a first feature amount, and a second extraction means for extracting a first feature amount using a second parameter different from the first parameter. The same as the image data processed using the first parameter The recording medium is characterized by comprising a second extraction means for processing the image data to extract a second feature amount, and a discrimination means for discriminating the type of the recording medium based on the first feature amount and the second feature amount. [Effects of the Invention]

[0009] According to the present invention, it is possible to reduce the time required to identify the recording medium. [Brief explanation of the drawings]

[0010] [Figure 1] 1 is a block diagram illustrating an example of a system configuration according to an embodiment of the present invention. [Figure 2] 1 is a block diagram showing an example of a system configuration according to an embodiment of the present invention; [Figure 3] FIG. 1 is a diagram illustrating an internal configuration of a recording apparatus according to an embodiment of the present invention. [Figure 4] FIG. 2 is a diagram illustrating an example of a sensor unit according to the present embodiment. [Figure 5] FIG. 2 is a block diagram illustrating an example of a functional configuration according to the present embodiment. [Figure 6] 10 is a flowchart of a paper type determination process according to the present embodiment. [Figure 7] 6 is a flowchart showing a process for determining the type of paper in the present embodiment. [Figure 8] 2 is a diagram illustrating the configuration of a convolution calculation unit and a paper type determination unit according to the present embodiment. FIG. [Figure 9] 1A and 1B are diagrams illustrating examples of an input image and an output image when image data is input in this embodiment. [Figure 10] FIG. 10 is a diagram illustrating pre-processing of image data in the present embodiment. [Figure 11] 6 is a flowchart showing a process for determining the type of paper in the present embodiment. DETAILED DESCRIPTION OF THE INVENTION

[0011] (First embodiment) Hereinafter, embodiments will be described in detail with reference to the accompanying drawings. Note that the following embodiments do not limit the scope of the claimed invention. Although multiple features are described in the embodiments, not all of these multiple features are necessarily essential to the invention, and multiple features may be combined arbitrarily. Furthermore, in the accompanying drawings, the same reference numerals are used to designate the same or similar components, and redundant explanations will be omitted.

[0012] (System Configuration) An example of a system configuration according to this embodiment will be described using Figures 1 and 2. In this embodiment, machine learning is performed to identify the paper type, and the paper type is identified using a trained estimation model obtained by machine learning. As an example of a system configuration for such processing, the example in Figure 1 shows an example in which machine learning and paper type identification are performed by a single device, and the example in Figure 2 shows an example in which machine learning and paper type identification are shared and performed by multiple devices. The medium to be identified is not limited to paper, and may be a recording medium such as cloth or a PVC sheet. In the following description, identification of paper type will be described as an example.

[0013] The configuration of the device in the system configuration of FIG. 1 will be described. For convenience, this device will be referred to as a paper type identification device below, but it may be configured to perform processes other than identifying paper types. In this embodiment, the paper type identification device is provided, for example, inside a recording device. The paper type identification device identifies the type of paper based on paper characteristic values ​​measured using a sensor unit included in the paper transport mechanism of the recording device, as described below. The paper type identification device includes, for example, a CPU 100, a paper type identification unit 101, a learning unit 102, a ROM 103, a RAM 104, a NVRAM 105, a display / operation unit 106, a controller 107, a LAN unit 108, and a network driver 109. ROM is an abbreviation for Read Only Memory, RAM is an abbreviation for Random Access Memory, and NVRAM is an abbreviation for Non-Volatile RAM. LAN is an abbreviation for Local Area Network. The paper type identification device may be provided separately from the recording device and capable of communicating with the recording device. The sensor unit may also be provided in the paper type identification device.

[0014] The paper type identification device executes various control processes, such as overall device control and paper type identification processing, by having the CPU 100 execute programs stored in the ROM 103 or the like. At this time, the RAM 104 can be used, for example, as a work memory during control. Data that should be retained when the paper type identification device is powered off is retained in the NVRAM 105, which is nonvolatile RAM. The paper type identification device controls the paper type identification unit 101 and learning unit 102 by having the CPU 100 execute the control programs stored in the ROM 103. In this case, the RAM 104 retains, for example, data resulting from paper measurements as temporary records. The NVRAM 105 stores records of various data necessary for maintaining the paper type identification device and information about paper used to identify paper types. The paper type identification unit 101 executes paper type identification processing and identifies the type of paper based on the data resulting from measurements of the paper. The learning unit 102 performs machine learning to obtain a trained estimation model to be used in the paper type identification processing. The operations of the paper type identification unit 101 and the learning unit 102 will be described later.

[0015] The paper type identification device displays information on a screen and accepts user operations via, for example, the display / operation unit 106. Information can be displayed not only via a screen display but also using various interfaces such as audio and vibration. User operations are accepted via hardware such as a keyboard, pointing device, or touchpad. The display / operation unit 106 may be implemented as separate hardware components such as a display and keyboard, or as a single piece of hardware such as a touch panel. The controller 107 converts information output by the CPU 100, for example, into a format usable by the display / operation unit 106 to generate information that can be presented to the user, and outputs the converted information to the display / operation unit 106. The controller 107 also converts user operations accepted by the display / operation unit 106 into a format that can be processed by the CPU 100, and outputs the converted information to the CPU 100. The execution and setting operations of each function in the paper type identification device are executed via, for example, the display / operation unit 106 and the controller 107.

[0016] The paper type identification device is also connected to a network via, for example, a LAN unit 108 and communicates with other devices. The network driver 109 extracts data to be handled by the CPU 100 from signals received via the LAN unit 108 and converts data output from the CPU 100 into a format for sending to the network. The LAN unit 108 may include an interface such as a socket for wired communication such as Ethernet (registered trademark) and a signal processing circuit. The LAN unit 108 may also include an antenna for wireless communication such as a wireless LAN conforming to the IEEE802.11 standard series and a signal processing circuit. Instead of the LAN unit 108, a communication unit for public wireless communication, short-range wireless communication, or the like may be provided. When the paper type identification device is operated via a remote user interface, control commands and setting value acquisition to the paper type identification device and processing results output may be performed via the LAN unit 108 and the network driver 109.

[0017] FIG. 2 shows an example of a system configuration in which the learning unit 102 is located outside the paper type identification device. In the example of FIG. 2, the system includes a paper type identification device 200 and a server 202. The paper type identification device 200 connects to the server 202 via a network 201, acquires a trained estimation model that is the result of machine learning in a learning unit 204 in the server 202, and executes paper type identification processing using the trained estimation model. The server 202 may be a processing device configured, for example, by a general-purpose computer having a CPU 203, a ROM 205, a RAM 206, an NVRAM 207, etc. The server 202 controls the machine learning processing in the learning unit 204 by, for example, causing the CPU 203 to execute a program stored in the ROM 205. The server 202 may also have a GPU, and use the GPU to control the machine learning processing in the learning unit 204. A GPU can perform efficient calculations by processing a large amount of data in parallel, so it is effective to use a GPU for processing when performing learning multiple times using a learning model such as deep learning. Specifically, when a learning program including a learning model is executed, the CPU 203 and GPU work together to perform calculations to perform learning. The processing of the learning unit 204 may be performed solely by the GPU. The estimation unit 506 and learning unit 503 described in FIG. 5 may also use a GPU, similar to the learning unit 204. The RAM 206 temporarily stores data used in learning during server control execution. The NVRAM 207 stores various data necessary for generating a trained estimation model and information about paper for identifying paper types. The server 202 may also have a LAN unit 208 for connecting to an external device, such as the paper type identification device 200. The LAN unit 208 and the network driver 209 located between the LAN unit 208 and the CPU 203 are similar to the LAN unit 108 and network driver 109 in FIG. 1. The following embodiment will be described using the system configuration of FIG. 1 as an example.

[0018] FIG. 3 is a diagram showing the schematic internal configuration of the recording apparatus according to the embodiment.

[0019] 3 is provided with a cassette paper feed unit 9 and a manual tray paper feed unit (not shown), and performs recording on paper fed from either paper feed unit. For example, when paper P is fed from cassette paper feed unit 9, it is fed and transported by pickup roller 10, and when it reaches the position of nip roller pair 4, nip roller pair 6 rotates and the paper is transported in the transport direction (leftward in the figure).

[0020] The paper P is sandwiched between the conveyor belt 8 and the pinch roller pair 5, and is conveyed in the conveying direction as the conveyor belt 8 moves. Then, the paper P conveyed along with the conveyor belt 8 is conveyed to the recording start position of the inkjet recording heads 1 (1C, 1M, 1Y, 1K).

[0021] Conveyor belt 8 is stretched by drive roller 6 and driven roller 7. In recording device 300, the position where paper P is nipped by pinch roller pair 5 is set as the recording start position, and an image is recorded at a predetermined position on the recording medium by controlling the recording timing of recording head 1 based on the position of drive roller 6. When recording is complete, the paper passes through paper discharge roller pair 11 and is discharged.

[0022] The recording head 1 has line heads of each color lined up along the transport direction and is attached to the recording head unit 2. Cyan ink is ejected from head 1C, magenta ink from head 1M, yellow ink from head 1Y, and black ink from head 1K. Each recording head 1 is supplied with ink via a tube from an ink tank (not shown) that stores the ink of each color independently.

[0023] The line head for each color may be formed with a single nozzle tip, or may be formed with divided nozzle tips arranged in a line or in a regular pattern such as a staggered arrangement. In this embodiment, the line head for each color is described as a so-called full multi-line head in which nozzles are arranged in a range that covers the width of the printing area of ​​the largest size sheet that can be used by the printing apparatus 300.

[0024] Further, on the upstream side of the recording head unit 2, there is provided a sensor unit 3 having an imaging section that captures an image of the printing surface.

[0025] FIG. 4 is a diagram showing an example of the sensor unit 3. As shown in FIG.

[0026] The sensor unit 3 includes an image sensor 301 , a light source 302 , and a lens 303 .

[0027] The image sensor 301 may be a line sensor with a one-dimensional array of light receiving elements or an area sensor with a two-dimensional array of light receiving elements, or may be a sensor using a CCD, CMOS, or the like as the light receiving elements.

[0028] The light source 302 may be an LED or a laser, but is not limited to the above, as long as it can irradiate light onto the paper P. The lens 303 is a lens for condensing the reflected light that is incident on the paper P.

[0029] Light source 302 irradiates the printing surface of paper P, and image sensor 301 captures the reflected light for a certain period of time. An image on the light-irradiated surface of paper P is formed based on the received light and converted into image data. In this example, image sensor 301 is configured to capture reflected light from light source 302, but light source 302 may also be located on the non-printing side of paper P and capture transmitted light. In this way, image sensor 301 captures an image of a predetermined area on the printing surface of paper P and obtains image data.

[0030] With the above configuration, it is possible to acquire the characteristics of the printing surface of the paper. In this embodiment, the printing surface is imaged to acquire image data, but it is also possible to acquire image data by imaging the non-printing surface.

[0031] (Machine Learning) In this embodiment, the paper type identification process is performed using estimated parameters obtained in advance by machine learning, which will be described later. An example of the functional configuration related to this process will now be described with reference to Fig. 5. The processing units in Fig. 5 are configured to be able to communicate with each other and are connected by a bus or the like. Components 401 to 405, which will be described below, are configured in the paper type identification unit 101, and components 406 and 407 are configured in the learning unit 102.

[0032] The image receiving unit 401 functions as an image data acquiring unit that acquires image data captured using the image sensor 301 of the sensor unit 3 .

[0033] The convolution calculation unit 403 extracts feature quantities such as the amount of unevenness of the paper, the spacing between the unevenness, and the depth of the unevenness from the image data acquired by the image receiving unit 401 using parameters stored in the estimated parameter storage unit 405. The feature quantities extracted by the convolution calculation unit 403 are input to a paper type discrimination unit (recording medium discrimination unit) 404, which discriminates the type of paper. Here, the parameters are parameters used by the convolution calculation unit 403 to perform image processing, such as filter coefficients and biases.

[0034] Information about the determined paper type is displayed on the display / operation unit 106. The user confirms or sets the final paper type using the UI of the display / operation unit 106. An operation information acquisition unit 407 acquires operation information from the display / operation unit 106.

[0035] The learning unit 406 updates parameters such as the filter coefficients and biases used in the convolution calculation unit 403 based on the type of paper estimated by the paper type discrimination unit 404 and the operation information acquired by the operation information acquisition unit 407, and stores the updated parameters in the estimated parameter memory unit 405.

[0036] 6 is a flowchart of the paper type discrimination process of the paper type identification unit 101 of this embodiment. This process is executed by the CPU 100 controlling each component in accordance with a program stored in the ROM 103. This process is started when the user sets paper in the recording device 300 and instructs the recording device to feed the set paper. When the paper feeding command is given, the paper is transported by the rollers and belts as described in FIG. 3.

[0037] When the sheet is conveyed to a position facing the sensor unit 3, the sensor unit 3 acquires image data of the printing surface (step S510).

[0038] In step S520, the paper type is determined based on the image data acquired in step S510. The determination calculates the probability that each paper type is the measured paper type. Details of paper type determination will be described with reference to FIG.

[0039] In step S530, the paper type with the highest probability from the determination results of step S520 is set as the paper type to be used, and is displayed on the display / operation unit 106. If the displayed paper type is acceptable to the user, the user inputs on the display / operation unit 106 that the currently displayed paper type is to be set. On the other hand, if the user wants to set a paper type different from the displayed paper type, the paper type to be set can be changed from the display / operation unit 106. Furthermore, if the paper type set in the print driver differs from the determination result, the setting is changed to the paper type with the highest probability from those determined in step S520. Note that the print driver setting may not be changed automatically, and the set paper type may be changed only when the user confirms via the UI whether to change the paper type and gives permission.

[0040] In step S540, it is determined whether to update the estimated parameters. The user can set in advance, using the results of steps S510 to S530, whether to update the estimated parameters, that is, whether to perform re-learning, on the display / operation unit 106. If the learning execution setting is NO, the paper type discrimination process ends.

[0041] On the other hand, if the learning execution setting is YES, the process proceeds to step S550. In step S550, operation information acquisition unit 407 acquires the type of recording medium finally selected by the user.

[0042] In step S560, the learning unit 406 performs learning using the finally set paper type and the result of the determination by the paper type determination unit 404 as input data, and the estimated parameters are updated. In step S570, the updated results are stored in the estimated parameter storage unit 405. The estimated parameters are updated by calculating the error gradient for each parameter of the convolution calculation unit 403 and the paper type determination unit 404 using the well-known error backpropagation method, based on the error between the probability output by the paper type determination unit 404 and the target output stored in advance in ROM 103.

[0043] In this way, the paper type determination process of FIG. 6 ends. Then, parameters related to the recording operation, such as the transport amount and ink ejection amount, are set for recording according to the type of recording medium determined by the determination process. While step S530 displayed one paper type with the highest probability, multiple paper types may be displayed for the user to select. The method for selecting multiple types may be to display a predetermined number of types in descending order of probability, or to display only paper types with a predetermined probability or higher. While the above process is configured to display the information on the display / operation unit 106, if the device has a speaker and microphone, the paper type may be notified by announcing the name of the paper type from the speaker, and the user may input the paper type they want to select through the microphone.

[0044] Next, step S520 in FIG. 6 will be described with reference to FIG.

[0045] First, in step S601, the convolution calculation unit 403 extracts feature amounts from the image data acquired in step S510. The extracted feature amounts are input to the paper type discrimination unit 404, which then performs a primary discrimination to determine whether the paper from which the image data was acquired belongs to group 1 or group 2 (S602). Here, the parameters of the convolution calculation unit 403 and the paper type discrimination unit 404 are set to parameters stored in the estimated parameter storage unit 405 in association with the group discrimination. A method for creating the parameters will be described later.

[0046] In this embodiment, the paper type is identified from among plain paper, matte paper, glossy paper 1 (Canon Photo Paper Glossy Standard, manufactured by Canon), glossy paper 2 (Canon Photo Paper Glossy Gold, manufactured by Canon), glossy paper 3 (Canon Photo Paper Platinum Grade, manufactured by Canon), and semi-glossy paper. Furthermore, in this embodiment, plain paper, matte paper, and glossy paper 1 belong to group 1, while glossy paper 2, glossy paper 3, and semi-glossy paper belong to group 2. Papers with similar characteristics are grouped together. In this embodiment, as will be described later in FIG. 9, to extract information about the paper's unevenness as a feature, papers with similar unevenness are grouped together. Here, group 1 is composed of papers with a lot of unevenness, while group 2 is composed of papers that are flatter than group 1.

[0047] The convolution calculation unit 403 and paper type discrimination unit 404 are configured as shown in Fig. 8. The convolution calculation unit 403 has convolution layers C1 and C2 and pooling layers P1 and P2, and executes feature extraction of input image data. The paper type discrimination unit 404 has fully connected neural networks F1 and F2, and inputs the calculation results of the convolution calculation unit 403 and outputs the probability for each type of recording medium. The neural network can be GAP (Global Average Pooling), etc.

[0048] FIG. 9 shows an input image and a portion of the output images of the convolution layers C1 and C2 and the pooling layers P1 and P2 when image data of matte paper and glossy paper are input to the convolution operation unit 403, respectively.

[0049] 8-1 to 8-5 represent the input image of matte paper and a portion of the output image of each layer when input to the convolution calculation unit 403, and 8-2 and 8-4 represent the results of the convolution calculation, from which feature amounts such as the amount of unevenness of the paper, the spacing between the unevenness, and the depth of the unevenness are extracted.

[0050] 8-3 and 8-5 perform MAX pooling, extracting only the pixel with the largest value from among the neighboring pixels, so even if the image becomes coarse, the parts from which features have been extracted are emphasized.

[0051] 8-6 to 8-10 show an input image of glossy paper and a portion of the output image of each layer when input to the convolution calculation unit 403. Even for glossy paper with little unevenness, feature amounts such as the depth and width of the unevenness are extracted.

[0052] Below are some examples of study methods and study conditions. First, an example of learning conditions is shown. The learning update rate is 0.0001, the filter size of C1 is 10 x 10, the number of filters in C1 is 30, the filter size of C2 is 5 x 5, the number of filters in C2 is 30, and the pooling size of P1 and P2 is 2 x 2. The pooling method is MAX pooling, and the number of hidden layers in the fully connected layer is 800. These conditions are somewhat empirically set values, and are determined by trial and error, changing the parameters until the target discrimination accuracy is achieved. As an example, in this embodiment, the filter size of C2 is 5 x 5, after comparing the results of 3 x 3, 5 x 5, and 10 x 10.

[0053] Next, regarding the learning method, for the primary classification in S602, 500 images of each of the papers to be classified (plain paper, matte paper, glossy paper 1, glossy paper 2, glossy paper 3, semi-glossy paper) are prepared, labeled with correct answers, and learning is performed. The correct labels are group 1 for plain paper, matte paper, and glossy paper 1, and group 2 for glossy paper 2, glossy paper 3, and semi-glossy paper. After learning, 100 images of each paper type are used for verification to verify the classification accuracy. If the obtained classification accuracy does not meet the target, the learning conditions mentioned above are changed and learning is performed again. For the secondary classification in S606, 500 images of each of the papers to be classified (plain paper, matte paper, glossy paper 1) are prepared, labeled with correct answers, and learning is performed. After learning, 100 images of each paper type are used for verification to verify the classification accuracy. Similarly, for the secondary classification in S610, 500 images of each of the papers to be classified (glossy paper 2, glossy paper 3, semi-glossy paper) are prepared, labeled with correct answers, and learning is performed. After learning, the accuracy of the classification is verified using images of 100 sheets of each paper for verification.

[0054] As an example, let us consider the parameters of the convolutional layer obtained as a result of learning. Since the main spatial frequencies (roughness of the paper surface) are different between Group 1 and Group 2, the parameters of the convolutional layer are values ​​learned to configure filters appropriate for the spatial frequencies of each group.

[0055] If the result of the primary discrimination in step S602 is group 1 (step S603), in step S604, estimated parameters are set for the convolution calculation unit 403 and the paper type discrimination unit 404. Here, the estimated parameters stored in the estimated parameter storage unit 405 corresponding to group 1 are set.

[0056] Next, in step S605, a convolution operation is performed on the image data acquired in step S510 using the estimation parameters set in step S604 to extract feature quantities.

[0057] In step S606, the feature amount extracted in step S605 is input to the paper type discrimination unit 404, and a secondary discrimination is performed to discriminate which paper belongs to group 1. This determines the detailed type of recording medium.

[0058] On the other hand, if the result of the primary discrimination in step S602 is group 2 (S607), in step S608, estimated parameters are set for the convolution calculation unit 403 and the paper type discrimination unit 404. Here, the estimated parameters are set to those stored in the estimated parameter storage unit 405 corresponding to group 2.

[0059] Next, in step S609, a convolution operation is performed on the image data acquired in step S510 using the estimation parameters set in step S608 to extract feature quantities.

[0060] In step S610, the feature amount extracted in step S609 is input to the paper type discrimination unit 404, and a secondary discrimination is performed to discriminate which paper belongs to group 2. This determines the detailed type of recording medium.

[0061] In this way, by extracting features from the same image data using different estimation parameters, it is possible to determine not only the general type but also the detailed type of recording medium, such as whether it is glossy paper 2 or glossy paper 3, without having to take multiple images under different conditions.

[0062] Although Fig. 6 illustrates the process of determining a unique type by performing two determinations, more determinations may be performed depending on the paper type to be determined and the number of types to be determined. This will be explained below with reference to Fig. 11.

[0063] 11, similar to step S601, the convolution calculation unit 403 extracts features from the image data acquired in step S510. In step S902, the extracted features are input to the paper type discrimination unit 404, which then performs a primary discrimination to discriminate the group of paper from which the image data was acquired.

[0064] In step S903, the estimated parameters of the convolution calculation unit 403 and the paper type discrimination unit 404 are set to the estimated parameters stored in the estimated parameter storage unit 405 corresponding to the group determined in step S902.

[0065] In step S904, a convolution operation is performed on the image data acquired in step S510 using the estimation parameters set in step S903 to extract feature quantities.

[0066] In step S905, the feature amount extracted in step S904 is input to the paper type discrimination unit 404, and it is determined which of group 1 the paper belongs to.

[0067] In step S906, it is determined whether the result of the determination in step S905 is a unique paper type. If there are multiple candidates and one paper type has not been determined, the process returns to step S903 and continues. If the paper type is a unique one, the process ends.

[0068] The method for determining the number of groups and the paper types that make up each group in this embodiment will be described below.

[0069] First, estimation parameters are created to be set in the convolution calculation unit 403 and the paper type discrimination unit 404, which discriminate all paper types to be discriminated into a unique type in one discrimination.

[0070] Next, new image data for each paper type not used when creating the estimated parameters is input to the convolution calculation unit 403, and the "correct paper type" and the "output paper type with a probability of 1% or more" are grouped together based on the probability of each paper type output by the paper type discrimination unit 404. For example, if the correct paper type is plain paper and the probability of it being matte paper is determined to be 3%, plain paper and matte paper will be grouped together as the same recording medium.

[0071] The method for creating the estimated parameters in this embodiment will be described below. The estimated parameters are created by using a data set created by the method described below as input data and performing learning using the backpropagation algorithm in the paper type identification unit 101 and the learning unit 102.

[0072] The method for creating the dataset is explained below.

[0073] 10 shows the preprocessing of the acquired image data, in which image data 91 is captured under appropriate exposure conditions only for the portion where light emitted by the light-emitting element is reflected, resulting in crushed black areas except for a central properly exposed portion 92. Therefore, only the properly exposed portion 92 is cut out as a perfect circle, and then cut out as a square 93 inscribed in the cut-out perfect circle.

[0074] Divide the square 93 inscribed in the cut-out circle into 6 equal parts both vertically and horizontally, and cut out a square consisting of the top left 4 squares.

[0075] Next, the image is cut out by sliding it horizontally one square at a time, and when it reaches the right edge, it returns to the left edge and slides downward one square.By again sliding it horizontally one square at a time and cutting it out, 25 pieces of image data 94 for creating estimation parameters are created from one piece of image data.

[0076] In this embodiment, the size of the image data used for parameter creation is a monochrome image of 186 x 186 pixels. Using the above method, image data for parameter creation is created for each type of paper to be identified, and the correct identification results are labeled to create a data set.

[0077] The estimation parameters used in step S602 only need to be able to determine which group it is, and in this case, it is sufficient to know whether there is a lot of unevenness or not. Therefore, in order to learn the differences in feature amounts between the groups, the image data of the paper sheets of both groups created by the above method is used for learning. The image data of the paper sheets of both groups is used as input data, and learning is performed using the groups (group 1, group 2) as correct labels. The parameters obtained by this learning are used as the estimation parameters to be used in step S602.

[0078] The estimation parameters set in step S604 need to identify which paper in group 1 it is. In order to learn the differences in the feature amounts of the papers within group 1, only the image data of group 1 is used for learning. Learning is carried out using the image data of group 1 as input data and the paper types within group 1 (plain paper, matte paper, glossy paper 1) as correct labels. The parameters obtained through this learning are used as the estimation parameters in step S604. The estimation parameters set in step S608 are also created in a similar manner using the image data of group 2.

[0079] In this way, learning is performed in advance to determine the estimated parameters. These estimated parameters are stored as initial values ​​in the estimated parameter storage unit 405. If the user has set the estimated parameters to be updated, additional learning is performed in step S560 of FIG. 6, and the estimated parameters are updated.

[0080] In this way, by determining the type of paper based on the captured image of the paper, it is possible to determine the type of paper in detail. [Explanation of symbols]

[0081] 100 CPU 101 Paper type identification unit 102 Study Units 106 Display / operation section 3 Sensor Unit

Claims

1. an image data acquisition means for acquiring image data obtained by imaging a predetermined area of ​​a recording medium; a first extraction means for processing the image data using a first parameter to extract a first feature amount; a second extraction means for extracting a second feature amount by processing the image data, which is the same as the image data processed using the first parameter, using a second parameter different from the first parameter; a discrimination means for discriminating the type of the recording medium based on the first characteristic amount and the second characteristic amount; A recording medium discrimination device comprising:

2. the determining means includes a first determining means and a second determining means, the first determination means determines, based on the first feature extracted by the first extraction means, which group the recording medium belongs to, among groups into which a plurality of types of recording media are divided based on features of image data; 2. The recording medium discrimination device according to claim 1, wherein the second discrimination means discriminates, based on the second feature extracted by the second extraction means, which type of the recording medium belongs to among the types of recording media belonging to the group discriminated by the first discrimination means.

3. a third extraction unit that processes the image data using a third parameter different from the first parameter and the second parameter to extract a third feature amount; The determining means includes a third determining means, The recording medium discrimination device according to claim 2, wherein the third discrimination means, when the second discrimination means does not determine one type of the candidate recording medium, discriminates, based on the third characteristic amount, which type of the candidate recording medium discriminated by the second discrimination means the recording medium belongs to.

4. 4. The recording medium discrimination device according to claim 2, further comprising a selection means for selecting the second parameter to be used in the second extraction means from a plurality of parameters set corresponding to each group based on the group discriminated by the first discrimination means.

5. 5. The recording medium discrimination device according to claim 2, wherein the first parameters are parameters created by learning such that when image data is input to the first extraction means using image data of a type of recording medium belonging to each group as input data and a group corresponding to the input data as a correct label, the group discriminated by the first discrimination means is the group of the correct label.

6. 6. The recording medium discrimination device according to claim 2, wherein the second parameter is a parameter created by learning such that when image data is input to the second extraction means using image data of a type of recording medium belonging to a predetermined group as input data and a type of recording medium corresponding to the input data as a correct label, the type of recording medium discriminated by the second discrimination means becomes the type of recording medium of the correct label.

7. 7. The recording medium discrimination device according to claim 1, wherein the first extraction means and the second extraction means extract the feature amounts by performing a convolution operation on the image data.

8. 8. The recording medium discrimination device according to claim 1, wherein the discrimination means includes a neural network.

9. a recording means for performing recording by ejecting ink onto the recording medium; 7. The recording medium discrimination device according to claim 2, wherein a parameter for ink ejection amount during recording is set in accordance with the type of recording medium discriminated by said second discrimination means.

10. having an image sensor, 7. The recording medium discrimination device according to claim 2, wherein the image data acquisition means acquires image data obtained by imaging the predetermined area of ​​the recording medium with the image sensor.

11. An image of a predetermined area of ​​the recording medium is taken, Acquire the captured image data, extracting a first feature amount from the image data using a first parameter; extracting a second feature amount from the image data that is the same as the image data processed using the first parameter, using a second parameter different from the first parameter; A method for identifying a recording medium, comprising: identifying the type of the recording medium based on the first characteristic amount and the second characteristic amount.

Images