TWI745068B - Method for establishing seal identification model and server terminal for establishing seal identification model - Google Patents

Method for establishing seal identification model and server terminal for establishing seal identification model Download PDF

Info

Publication number
TWI745068B
TWI745068B TW109130059A TW109130059A TWI745068B TW I745068 B TWI745068 B TW I745068B TW 109130059 A TW109130059 A TW 109130059A TW 109130059 A TW109130059 A TW 109130059A TW I745068 B TWI745068 B TW I745068B
Authority
TW
Taiwan
Prior art keywords
image
seal
server
seal image
training
Prior art date
Application number
TW109130059A
Other languages
Chinese (zh)
Other versions
TW202211111A (en
Inventor
劉邦旭
邱建中
李藝鋒
宋政隆
王俊權
Original Assignee
中國信託商業銀行股份有限公司
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by 中國信託商業銀行股份有限公司 filed Critical 中國信託商業銀行股份有限公司
Priority to TW109130059A priority Critical patent/TWI745068B/en
Application granted granted Critical
Publication of TWI745068B publication Critical patent/TWI745068B/en
Publication of TW202211111A publication Critical patent/TW202211111A/en

Links

Images

Landscapes

  • Image Analysis (AREA)
  • Collating Specific Patterns (AREA)

Abstract

一種印鑑辨識模型建立方法，藉由一伺服端實施，該伺服端對於所儲存的多張印鑑影像之每一者進行色彩變換處理以獲得一異色印鑑影像，並自所儲存的多張表單影像及該等異色印鑑影像中分別選取一目標表單影像及一目標印鑑影像，將該目標印鑑影像疊合於該目標表單影像上以產生一用印影像，且根據該用印影像產生一指示出該用印影像中該目標印鑑影像所對應之位置的標記位置資料，並產生一包括該用印影像及該標記位置資料的訓練資料，以及根據所有訓練資料利用一物件偵測演算法建立一印鑑辨識模型。A method for establishing a seal recognition model is implemented by a server. The server performs color conversion processing on each of a plurality of stored seal images to obtain a different-color seal image, and obtains a different-color seal image from the stored multiple form images and A target form image and a target seal image are selected from the different-color seal images, the target seal image is superimposed on the target form image to generate a print image, and an indication of the print image is generated based on the print image The mark location data of the location corresponding to the target seal image is generated, and training data including the seal image and the mark location data are generated, and an object detection algorithm is used to establish a seal recognition model based on all the training data.

Description

印鑑辨識模型建立方法及用於建立印鑑辨識模型的伺服端Method for establishing seal identification model and server terminal for establishing seal identification model

本發明是有關於一種辦公室自動化方法，特別是指一種印鑑辨識模型的建立方法。The invention relates to an office automation method, in particular to a method for establishing a seal identification model.

電子文件已普遍而深入地應用於各行各業中，其中，關於例如銀行業、保險業，特別需要於電子文件中準確地辨別出印鑑影像，以利進行後續例如驗證等手續的需要。Electronic documents have been widely and intensively used in various industries. Among them, for example, banking and insurance industries, it is particularly necessary to accurately identify seal images in electronic documents to facilitate subsequent procedures such as verification.

目前，關於自動辨識電子文件中之印鑑影像的方式，是先建立一個有效的印鑑辨識模型，之後，才能將待辨識的電子文件投入該印鑑辨識模型中，用以自動地辨識出該待辨識的電子文件的印鑑圖案。而關於建立印鑑辨識模型的過程，則是藉由設定者的輸入操作產生包含多個相關於顏色辨識的規則，以供該印鑑辨識模型在收到一張待辨識的影像時，能夠根據顏色辨識的規則進行辨識，舉例來說，該印鑑辨識模型收到待辨識的影像時，將影像中所有像素的像素值投射至RGB色彩空間，再判斷是否存在多個像素值中的R值、G值、B值皆符合預設門檻值的目標像素，當判斷存在有目標像素時，該印鑑辨識模型即將目標像素所構成的圖像作為所辨識出的印鑑影像。At present, the method of automatically recognizing the seal image in an electronic document is to first establish an effective seal recognition model, and then put the electronic document to be recognized into the seal recognition model to automatically identify the to-be-recognized seal image. The seal pattern of the electronic file. Regarding the process of establishing a seal recognition model, the input operation of the setter generates a number of rules related to color recognition, so that the seal recognition model can recognize according to the color when it receives an image to be recognized. For example, when the seal recognition model receives the image to be recognized, it projects the pixel values of all pixels in the image to the RGB color space, and then determines whether there are R values and G values among multiple pixel values , B values are all target pixels that meet the preset threshold value. When it is determined that there are target pixels, the seal recognition model takes the image formed by the target pixels as the recognized seal image.

就相關業者而言，依據上述建立印鑑辨識模型的過程所建立出的印鑑辨識模型，確實可以辨識出待辨識的電子文件中的印鑑影像，不過仍具有缺陷，詳細地說，上述的印鑑辨識模型能夠很好地辨識出紅色的印鑑影像，然而現實生活中的戳印五花八門，顏色也不盡相同，當該待辨識影像中的另一印鑑影像不為紅色(例如藍色、黑色等其他顏色)時，現有印鑑模型則無法辨識出該另一印鑑影像。As far as the relevant industry is concerned, the seal identification model established based on the above process of establishing the seal identification model can indeed identify the seal image in the electronic document to be identified, but it still has defects. In detail, the above-mentioned seal identification model The red seal image can be recognized very well. However, in real life, there are many kinds of stamps with different colors. When another seal image in the image to be recognized is not red (such as blue, black and other colors) At this time, the existing seal model cannot identify the other seal image.

但基於銀行業、保險業對於在電子文件中準確地辨別出印鑑影像，以利進行後續例如驗證工作等有更高要求的業者來說，更精準的印鑑辨識模型的建立過程，仍需要相關業者不斷的研究改進，進而滿足工作上的特殊需求。However, for the banking and insurance industries to accurately identify seal images in electronic documents to facilitate follow-up such as verification work, the establishment process of a more accurate seal recognition model still needs relevant industry players. Continuous research and improvement to meet the special needs of work.

因此，本發明的目的，即在提供一種建立能夠辨識不同顏色之印鑑影像之印鑑辨識模型的印鑑辨識模型建立方法。Therefore, the purpose of the present invention is to provide a seal recognition model establishment method capable of recognizing seal images of different colors.

再者，本發明的另一目的，在於提供一種建立能夠辨識不同顏色之印鑑影像之印鑑辨識模型的伺服端。Furthermore, another object of the present invention is to provide a server for establishing a seal recognition model capable of recognizing seal images of different colors.

於是，本發明印鑑辨識模型建立方法，藉由一連接一使用端的伺服端來實施，該印鑑辨識模型建立方法包含一步驟(A)、一步驟(B)、一步驟(C)、一步驟(D)，及一步驟(E)。Therefore, the method for establishing a seal recognition model of the present invention is implemented by a server connected to a user end. The method for establishing a seal recognition model includes one step (A), one step (B), one step (C), and one step ( D), and a step (E).

在該步驟(A)中，藉由該伺服端，對於該伺服端所儲存的多張相關於多個戳印的印鑑影像之每一者，將該印鑑影像進行色彩變換處理，以獲得一與處理前之該印鑑影像具相異顏色的異色印鑑影像。In this step (A), by the server, for each of the plurality of seal images related to the plurality of stamps stored by the server, the seal image is subjected to color conversion processing to obtain a and The seal image before processing is a different-color seal image with different colors.

在該步驟(B)中，藉由該伺服端，自該伺服端所儲存的多張相關於多張表單的表單影像及該等異色印鑑影像中分別選取一目標表單影像及一目標印鑑影像，並將該目標印鑑影像疊合於該目標表單影像之上以產生一用印影像。In this step (B), by the server, a target form image and a target seal image are selected from the multiple form images related to multiple forms and the different-color seal images stored in the server, respectively, And superimpose the target seal image on the target form image to generate a print image.

在該步驟(C)中，藉由該伺服端，根據該用印影像產生一指示出該用印影像中該目標印鑑影像所對應之位置的標記位置資料。In the step (C), the server generates a mark position data indicating the position corresponding to the target seal image in the print image based on the print image.

在該步驟(D)中，藉由該伺服端，產生一包括該用印影像及對應該用印影像之該標記位置資料的訓練資料。In the step (D), the server generates a training data including the printing image and the mark position data corresponding to the printing image.

在該步驟(E)中，藉由該伺服端，根據重複步驟(B)、(C)與(D)所產生的所有訓練資料，利用一物件偵測演算法建立一用於辨識出一待辨識影像中對應於一戳印物件之物件區域的印鑑辨識模型。In this step (E), with the server, based on all the training data generated by repeating steps (B), (C), and (D), an object detection algorithm is used to create an object detection algorithm to identify a waiting Identify the seal recognition model of the object area corresponding to a stamped object in the image.

再者，本發明用於建立印鑑辨識模型的伺服端，包含一伺服端儲存模組，及一伺服端處理模組。Furthermore, the present invention is used to establish the server side of the seal identification model, including a server side storage module and a server side processing module.

該伺服端儲存模組儲存有多張相關於多個戳印的印鑑影像，以及多張相關於多張表單的表單影像。The server-side storage module stores multiple seal images related to multiple stamps, and multiple form images related to multiple forms.

該伺服端處理模組電連接該伺服端儲存模組。The server-side processing module is electrically connected to the server-side storage module.

其中，該伺服端處理模組對該等印鑑影像之每一者，將該印鑑影像進行色彩變換處理，以獲得一與處理前之該印鑑影像具相異顏色的異色印鑑影像，並自該等表單影像及該等異色印鑑影像中分別選取一目標表單影像及一目標印鑑影像，並將該目標印鑑影像疊合於該目標表單影像之上以產生一用印影像，且根據該用印影像產生一指示出該用印影像中該目標印鑑影像所對應之位置的標記位置資料，並產生一包括該用印影像及對應該用印影像之該標記位置資料的訓練資料，以及根據重複前述過程所產生的所有訓練資料，利用一物件偵測演算法建立一用於辨識出一待辨識影像中對應於一戳印物件之物件區域的印鑑辨識模型。Wherein, the server-side processing module performs color conversion processing on the seal image for each of the seal images to obtain a different color seal image with a different color from the seal image before processing, and collects A target form image and a target seal image are respectively selected from the form image and the different-color seal images, and the target seal image is superimposed on the target form image to generate a print image, and an instruction is generated based on the print image Get the mark position data of the position corresponding to the target seal image in the imprint image, and generate a training data including the imprint image and the mark position data corresponding to the imprint image, and all training data generated by repeating the foregoing process Using an object detection algorithm to establish a seal recognition model for recognizing an object area corresponding to a stamped object in a to-be-recognized image.

本發明的功效在於：藉由該伺服端所產生的該等異色印鑑影像，進而產生出具有該目標印鑑影像之該用印影像以及包括該用印影像的該訓練資料，再根據所有訓練資料利用該物件偵測演算法建立該印鑑辨識模型，藉此，由於該等異色印鑑影像與處理前的該等印鑑影像具有不同的顏色，進而讓該伺服端所建立的該印鑑辨識模型能夠辨識不同顏色的印鑑影像。The effect of the present invention is that the different-color seal images generated by the server are used to generate the print image with the target seal image and the training data including the print image, and then use the object according to all the training data The detection algorithm establishes the seal identification model, whereby the different-color seal images and the seal images before processing have different colors, so that the seal identification model established by the server can identify seals of different colors image.

在本發明被詳細描述之前，應當注意在以下的說明內容中，類似的元件是以相同的編號來表示。Before the present invention is described in detail, it should be noted that in the following description, similar elements are denoted by the same numbers.

首先參閱圖1，本發明印鑑辨識模型建立方法的一實施例，由如圖1所示的一印鑑辨識模型建立系統來實施，該印鑑辨識模型建立系統包含本發明用於建立印鑑辨識模型的伺服端1，及一經由一通訊網路100連接至該伺服端1的使用端2。First, referring to FIG. 1, an embodiment of the method for establishing a seal recognition model of the present invention is implemented by a seal recognition model establishment system as shown in FIG. Terminal 1, and a user terminal 2 connected to the server terminal 1 via a communication network 100.

該伺服端1包含一伺服端通訊模組11、一伺服端儲存模組12，及一電連接該伺服端通訊模組11及該伺服端儲存模組12的伺服端處理模組13，該伺服端通訊模組11連接至該通訊網路100，該伺服端儲存模組12儲存有多張相關於多個戳印的印鑑影像、多張相關於多張表單的表單影像、多筆姓名，以及一用於根據一待處理姓名產生一對應該待處理姓名的模擬戳印圖像的印鑑影像產生器，例如MakePic.com網站所提供的線上***產生器，或是類似於Adobe公司的Adobe Illustraor的繪圖軟體。在此，該伺服端1是例如雲端伺服器、超級電腦、個人電腦，或是其他類似裝置其中任一。The server 1 includes a server communication module 11, a server storage module 12, and a server processing module 13 electrically connected to the server communication module 11 and the server storage module 12. The server The terminal communication module 11 is connected to the communication network 100. The server storage module 12 stores a plurality of seal images related to a plurality of stamps, a plurality of form images related to a plurality of forms, a plurality of names, and a A seal image generator used to generate a simulated stamp image of the name to be processed based on a name to be processed, such as the online seal generator provided by MakePic.com, or a drawing similar to Adobe Illustraor from Adobe software. Here, the server 1 is, for example, a cloud server, a super computer, a personal computer, or any other similar device.

該使用端2由一使用者所持有，並包含一使用端通訊模組21、一使用端輸入模組22，及一電連接該使用端通訊模組21及該使用端輸入模組22的使用端處理模組23，該使用端通訊模組21連接至該通訊網路100，該使用端輸入模組22用以供該使用者進行輸入操作。在此，該使用端2是例如個人電腦、平板電腦、筆記型電腦，或其他類似裝置其中任一。The user end 2 is held by a user, and includes a user end communication module 21, a user end input module 22, and an electrical connection between the user end communication module 21 and the user end input module 22 The user-side processing module 23, the user-side communication module 21 is connected to the communication network 100, and the user-side input module 22 is used for the user to perform input operations. Here, the user terminal 2 is, for example, any one of a personal computer, a tablet computer, a notebook computer, or other similar devices.

參閱圖1與圖2，本發明印鑑辨識模型建立方法的該實施例包含一步驟31、一步驟32、一步驟33、一步驟34、一步驟35，及一步驟36，說明如何建立一用於辨識出一待辨識影像中對應於一戳印物件之物件區域的印鑑辨識模型。1 and 2, this embodiment of the method for establishing a seal recognition model of the present invention includes a step 31, a step 32, a step 33, a step 34, a step 35, and a step 36, explaining how to create a A seal recognition model corresponding to an object area of a stamped object in a to-be-recognized image is recognized.

首先在該步驟31中，該伺服端處理模組13對於該伺服端儲存模組12所儲存的該等相關於多個戳印的印鑑影像之每一者，將該印鑑影像進行色彩變換處理，以獲得一與處理前之該印鑑影像具相異顏色的異色印鑑影像。其中，該伺服端儲存模組12所儲存的每一印鑑影像屬於一原生印鑑影像及一模擬印鑑影像之其中一者。屬於該原生印鑑影像的印鑑影像係該伺服端處理模組13根據一相關於一***的戳印圖像所產生的。另一方面，屬於該模擬印鑑影像的印鑑影像係該伺服端處理模組13根據該等姓名中的一目標姓名，利用該印鑑影像產生器產生之對應該目標姓名的模擬戳印圖像所對應產生的。First, in the step 31, the server-side processing module 13 performs color conversion processing on each of the seal images related to a plurality of stamps stored in the server-side storage module 12. To obtain a different-color seal image with a different color from the seal image before processing. Wherein, each seal image stored in the server storage module 12 belongs to one of a native seal image and an analog seal image. The seal image belonging to the native seal image is generated by the server-side processing module 13 based on a stamp image related to a seal. On the other hand, the seal image belonging to the simulated seal image corresponds to the simulated seal image corresponding to the target name generated by the seal image generator according to a target name among the names by the server-side processing module 13 produced.

參閱圖1與圖3，詳細而言，該步驟31包括一子步驟311、一子步驟312，及一子步驟313，說明該伺服端處理模組13如何對於該等印鑑影像進行色彩變換處理。1 and 3, in detail, the step 31 includes a sub-step 311, a sub-step 312, and a sub-step 313, explaining how the server-side processing module 13 performs color conversion processing on the seal images.

首先，在該步驟311中，對於每一印鑑影像，該伺服端處理模組13判斷出該印鑑影像是否為該模擬印鑑影像，當該伺服端處理模組13判斷出該印鑑影像為該模擬印鑑影像時，該伺服端處理模組13進行該子步驟312，另一方面，當該伺服端處理模組13判斷出該印鑑影像為該原生印鑑影像時，則該伺服端處理模組13將會直接進行該子步驟313。First, in step 311, for each seal image, the server-side processing module 13 determines whether the seal image is the analog seal image, and when the server-side processing module 13 determines that the seal image is the analog seal image The server-side processing module 13 performs the sub-step 312 during image processing. On the other hand, when the server-side processing module 13 determines that the seal image is the original seal image, the server-side processing module 13 will Proceed directly to this sub-step 313.

之後，在該子步驟312中，該伺服端處理模組13對該印鑑影像進行影像膨脹或影像侵蝕之其中一種的影像處理。其中，該伺服端處理模組13對該印鑑影像進行影像膨脹的影像處理，是為了模擬真實生活中人們在使用***蓋印時，***沾取過多印泥而產生的戳印，另一方面，該伺服端處理模組13對該印鑑影像進行影像侵蝕的影像處理，則是為了模擬真實生活中人們在使用***蓋印時，***沾取的印泥不足而產生的戳印。After that, in the sub-step 312, the server-side processing module 13 performs image processing of one of image expansion or image erosion on the seal image. Wherein, the server-side processing module 13 performs image expansion image processing on the seal image, in order to simulate the stamping caused by too much ink on the seal when people use the seal for stamping in real life. On the other hand, the The server-side processing module 13 performs image erosion processing on the seal image in order to simulate the stamping caused by insufficient ink on the seal when people use the seal for stamping in real life.

接著，在該子步驟313中，該伺服端處理模組13對屬於該原生印鑑影像的每一印鑑影像，及每一進行過影像膨脹或影像侵蝕之其中一種影像處理的印鑑影像，進行色彩變換處理，以獲得與處理前之該印鑑影像具相異顏色的該異色印鑑影像。Then, in the sub-step 313, the server-side processing module 13 performs color conversion on each seal image belonging to the native seal image and each seal image that has undergone one of image expansion or image erosion. Processing to obtain the different-color seal image with a different color from the seal image before processing.

再參閱圖2，之後，在該步驟32中，該伺服端處理模組13自該伺服端儲存模組12所儲存的多張相關於多張表單的表單影像及該等異色印鑑影像中，分別選取一目標表單影像及一目標印鑑影像，並將該目標印鑑影像疊合於該目標表單影像之上以產生一用印影像，值得一提的是，在日常生活中，一份正式文書不只在用印區會有戳印，當一書寫者在填寫資料因為誤繕而塗抹修改時，該書寫者也會在塗抹修改處蓋下戳印以證明其所填寫的資料並非由他人竄改，因此，在該步驟32中，該伺服端處理模組13係將該目標印鑑影像疊合於該目標表單影像之上的任意一處，以產生該用印影像。2 again, afterwards, in the step 32, the server processing module 13 stores the multiple form images related to multiple forms and the different-color stamp images stored by the server storage module 12, respectively Select a target form image and a target seal image, and superimpose the target seal image on the target form image to generate a print image. It is worth mentioning that in daily life, a formal document is not only used for print There will be a stamp in the area. When a writer fills in the information and changes it due to mistakes, the writer will also stamp the correction place to prove that the information he filled in was not tampered with by others. Therefore, in the In step 32, the server processing module 13 superimposes the target seal image on any place on the target form image to generate the print image.

接著，在該步驟33中，該使用端處理模組23根據該使用者藉由該使用端輸入模組22的輸入操作，產生一相關於該用印影像的標記請求，並透過該使用端通訊模組21經由該通訊網路100傳送至該伺服端1。Then, in the step 33, the user-side processing module 23 generates a marking request related to the printing image according to the input operation of the user through the user-side input module 22, and transmits it through the user-side communication module. The group 21 is transmitted to the server 1 via the communication network 100.

在該步驟34中，當該伺服端處理模組13經由該伺服端通訊模組11接收到來自該使用端2的該標記請求後，該伺服端處理模組13根據該標記請求產生一指示出該用印影像中該目標印鑑影像所對應之位置的標記位置資料。需要補充說明的是，在本實施例中，該步驟34是該伺服端處理模組13根據來自該使用端2的該標記請求產生該標記位置資料，但在其他實施例中，該伺服端1亦可僅包含該伺服端儲存模組12及該伺服端處理模組13，該伺服端處理模組13則直接根據該用印影像產生該標記位置資料，換言之，該伺服端處理模組13在進行該步驟32後，略過該步驟33直接進行該步驟34，而在該步驟34中，該伺服端處理模組13是根據該用印影像產生該標記位置資料，並不需要透過來自該使用端2的該標記請求產生該標記位置資料。In step 34, after the server-side processing module 13 receives the marking request from the user terminal 2 via the server-side communication module 11, the server-side processing module 13 generates an indication according to the marking request Mark location data of the location corresponding to the target seal image in the imprint image. It should be added that, in this embodiment, in step 34, the server processing module 13 generates the mark location data according to the mark request from the user terminal 2, but in other embodiments, the server terminal 1 It can also only include the server-side storage module 12 and the server-side processing module 13. The server-side processing module 13 directly generates the mark position data based on the printed image. In other words, the server-side processing module 13 is performing After the step 32, skip the step 33 and proceed directly to the step 34. In the step 34, the server-side processing module 13 generates the mark position data based on the printing image, and does not need to pass from the user terminal 2 The tag requested to generate the tag location data.

之後，在該步驟35中，該伺服端處理模組13產生一包括該用印影像及對應該用印影像之該標記位置資料的訓練資料。而當不斷地重複該步驟32、該步驟33、該步驟34，及該步驟35時，即可由該伺服端處理模組13建立出大量的訓練資料用以建立該印鑑辨識模型。After that, in step 35, the server-side processing module 13 generates a training data including the printing image and the mark position data corresponding to the printing image. When the step 32, the step 33, the step 34, and the step 35 are continuously repeated, a large amount of training data can be created by the server processing module 13 to build the seal recognition model.

接著，在該步驟36中，該伺服端處理模組13根據所有訓練資料，利用一物件偵測演算法，例如快速區域卷積神經網路(Faster Region-based Convolutional Neural Networks, Faster R-CNN)、單次多框偵測器(Single Shot Multibox Detector,SSD)，或是YOLO演算法(You Only Look Once)，建立用於辨識出一待辨識影像中對應於一戳印物件之物件區域的該印鑑辨識模型。詳細地說，該步驟36包括一子步驟361、一子步驟362、一子步驟363、一子步驟364，及一子步驟365，說明該伺服端處理模組13如何根據所有的訓練資料，建立該印鑑辨識模型。Then, in step 36, the server-side processing module 13 uses an object detection algorithm, such as Faster Region-based Convolutional Neural Networks (Faster Region-based Convolutional Neural Networks, Faster R-CNN) based on all training data. , Single Shot Multibox Detector (SSD), or YOLO algorithm (You Only Look Once), which is used to identify the object area corresponding to a stamped object in a to-be-recognized image Seal identification model. In detail, the step 36 includes a sub-step 361, a sub-step 362, a sub-step 363, a sub-step 364, and a sub-step 365, explaining how the server-side processing module 13 creates The seal identification model.

參閱圖1與圖4，首先，在該子步驟361中，該伺服端處理模組13將所有訓練資料分為一訓練子集及一測試子集。1 and 4, first, in the sub-step 361, the server processing module 13 divides all training data into a training subset and a test subset.

之後，在該子步驟362中，該伺服端處理模組13根據該訓練子集中的每一訓練資料中的該用印影像及該標記位置資料，利用該物件偵測演算法，例如但不限於前述的Faster R-CNN、SSD，或是YOLO演算法，建立用於辨識出一待辨識影像中對應於一戳印物件之物件區域的一訓練模型。Then, in the sub-step 362, the server-side processing module 13 uses the object detection algorithm, such as but not limited to the foregoing, according to the print image and the mark position data in each training data in the training subset The Faster R-CNN, SSD, or YOLO algorithm is used to establish a training model for recognizing an object area corresponding to a stamped object in a to-be-recognized image.

接著，在該子步驟363中，該伺服端處理模組13根據該測試子集中的每一訓練資料中的該用印影像及該標記位置資料，判斷出該訓練模型的辨識正確率是否高於一門檻值，當該伺服端處理模組13判斷出該訓練模型的辨識正確率並未高於該門檻值時，該伺服端處理模組13進行該子步驟364，亦即調整該訓練模型並重新回到該子步驟363，相反地，當該伺服端處理模組13判斷出該訓練模型的辨識正確率高於該門檻值時，該伺服端處理模組13進行該子步驟365，亦即確認該訓練模型為該印鑑辨識模型。Then, in the sub-step 363, the server processing module 13 determines whether the recognition accuracy of the training model is higher than a value based on the print image and the mark position data in each training data in the test subset. Threshold value. When the server-side processing module 13 determines that the recognition accuracy of the training model is not higher than the threshold value, the server-side processing module 13 performs the sub-step 364, that is, adjusts the training model and restarts Returning to the sub-step 363, on the contrary, when the server-side processing module 13 determines that the recognition accuracy of the training model is higher than the threshold value, the server-side processing module 13 performs the sub-step 365, that is, confirms The training model is the seal identification model.

綜上所述，本發明印鑑辨識模型建立方法，藉由該伺服端1產生該等異色印鑑影像，並根據所選取的該目標表單影像及該目標印鑑影像產生該用印影像，進而產生對應該用印影像的該標記位置資料，以及包括該用印影像及該標記位置資料的該訓練資料，最後再根據所有訓練資料利用該物件偵測演算法建立該印鑑辨識模型，藉此，由於該等異色印鑑影像與處理前的該等印鑑影像具有不同的顏色，因此該伺服端1判斷出該訓練模型的辨識正確率高於該門檻值時，代表所建立出的該印鑑辨識模型(即為該訓練模型)能夠辨識出不同顏色的各種印鑑影像，亦即當該待辨識影像中對應於該戳印物件之物件區域的顏色不為紅色或藍色等某一種預設顏色時，該印鑑辨識模型亦可偵測出對應於該戳印物件之物件區域，故確實能達成本發明的目的。In summary, the method for establishing a seal recognition model of the present invention generates the different-color seal images by the server 1, and generates the seal image based on the selected target form image and the target seal image, and then generates the corresponding seal image. The marking location data of the image and the training data including the printing image and the marking location data. Finally, based on all the training data, the object detection algorithm is used to establish the seal recognition model, thereby, due to the different-color seal images The seal images before processing have different colors. Therefore, when the server 1 determines that the recognition accuracy of the training model is higher than the threshold value, it represents the established seal recognition model (that is, the training model) It can recognize various seal images of different colors, that is, when the color of the object area corresponding to the stamped object in the image to be recognized is not a preset color such as red or blue, the seal recognition model can also detect The object area corresponding to the stamped object is measured, so it can indeed achieve the purpose of the invention.

惟以上所述者，僅為本發明的實施例而已，當不能以此限定本發明實施的範圍，凡是依本發明申請專利範圍及專利說明書內容所作的簡單的等效變化與修飾，皆仍屬本發明專利涵蓋的範圍內。However, the above are only examples of the present invention. When the scope of implementation of the present invention cannot be limited by this, all simple equivalent changes and modifications made in accordance with the scope of the patent application of the present invention and the content of the patent specification still belong to Within the scope covered by the patent of the present invention.

1:伺服端 100:通訊網路 11:伺服端通訊模組 12:伺服端儲存模組 13:伺服端處理模組 2:使用端 21:使用端通訊模組 22:使用端輸入模組 23:使用端處理模組 31~36:步驟 311~313:子步驟 361~365:子步驟 1: Server 100: Communication network 11: Server communication module 12: Server-side storage module 13: Server-side processing module 2: use end 21: User-side communication module 22: User input module 23: User-side processing module 31~36: Steps 311~313: Sub-step 361~365: substeps

本發明的其他的特徵及功效，將於參照圖式的實施方式中清楚地呈現，其中： 圖1是一方塊圖，說明實施本發明印鑑辨識模型建立方法的一實施例的一印鑑辨識模型建立系統； 圖2是一流程圖，說明本發明印鑑辨識模型建立方法的該實施例； 圖3是一流程圖，說明本發明印鑑辨識模型建立方法的該實施例中，一步驟31的子步驟；及 圖4是一流程圖，說明本發明印鑑辨識模型建立方法的該實施例中，一步驟36的子步驟。 Other features and effects of the present invention will be clearly presented in the embodiments with reference to the drawings, in which: Fig. 1 is a block diagram illustrating a seal recognition model establishment system implementing an embodiment of the seal recognition model establishment method of the present invention; Figure 2 is a flowchart illustrating this embodiment of the method for establishing a seal identification model of the present invention; Fig. 3 is a flowchart illustrating the sub-steps of step 31 in this embodiment of the method for establishing a seal identification model of the present invention; and FIG. 4 is a flowchart illustrating the sub-steps of step 36 in this embodiment of the method for establishing a seal identification model of the present invention.

31~36:步驟 31~36: Steps

Claims (6)

一種印鑑辨識模型建立方法，藉由一伺服端來實施，該伺服端還儲存有多筆姓名及一用於根據一待處理姓名產生一對應該待處理姓名的模擬戳印圖像的印鑑影像產生器，該印鑑辨識模型建立方法包含以下步驟：(A)藉由該伺服端，對於該伺服端所儲存的多張相關於多個戳印的印鑑影像之每一者，將該印鑑影像進行色彩變換處理，以獲得一與處理前之該印鑑影像具相異顏色的異色印鑑影像，其中，該伺服端所儲存的每一印鑑影像屬於一原生印鑑影像及一模擬印鑑影像之其中一者，屬於該原生印鑑影像的印鑑影像係該伺服端根據一相關於一***的戳印圖像所產生的，屬於該模擬印鑑影像的印鑑影像係該伺服端根據該等姓名中的一目標姓名，利用該印鑑影像產生器產生之對應該目標姓名的模擬戳印圖像所對應產生的；(B)藉由該伺服端，自該伺服端所儲存的多張相關於多張表單的表單影像及該等異色印鑑影像中分別選取一目標表單影像及一目標印鑑影像，並將該目標印鑑影像疊合於該目標表單影像之上以產生一用印影像；(C)藉由該伺服端，根據該用印影像產生一指示出該用印影像中該目標印鑑影像所對應之位置的標記位置資料；(D)藉由該伺服端，產生一包括該用印影像及對應該用印影像之該標記位置資料的訓練資料；及 (E)藉由該伺服端，根據重複步驟(B)、(C)與(D)所產生的所有訓練資料，利用一物件偵測演算法建立一用於辨識出一待辨識影像中對應於一戳印物件之物件區域的印鑑辨識模型。 A method for establishing a seal recognition model is implemented by a server that also stores multiple names and a seal image generation for generating a simulated stamp image corresponding to the name to be processed based on a name to be processed The method for establishing the seal recognition model includes the following steps: (A) By the server, for each of the plurality of seal images stored by the server, the seal image is color-coded Transformation processing to obtain a different-color seal image with a different color from the seal image before processing, wherein each seal image stored by the server belongs to one of a native seal image and an analog seal image, belonging to The seal image of the original seal image is generated by the server based on a seal image related to a seal, and the seal image belonging to the analog seal image is the server based on a target name among the names. The imitation stamp image corresponding to the target name generated by the seal image generator; (B) through the server, multiple form images related to multiple forms stored in the server and these A target form image and a target seal image are selected from the different-color seal image, and the target seal image is superimposed on the target form image to generate a print image; (C) by the server, according to the print image Generate a mark position data indicating the position corresponding to the target seal image in the use image; (D) generate a training data including the use image and the mark position data corresponding to the use image through the server; and (E) With the server, based on all the training data generated by repeating steps (B), (C), and (D), an object detection algorithm is used to create an object detection algorithm for identifying the image corresponding to A stamp recognition model of the object area of the stamped object. 如請求項1所述的印鑑辨識模型建立方法，在該步驟(A)中還包含以下子步驟：(A-1)對於每一印鑑影像，藉由該伺服端，判斷出該印鑑影像是否為該模擬印鑑影像；(A-2)當該伺服端判斷出該印鑑影像為該模擬印鑑影像時，藉由該伺服端，對該印鑑影像進行影像膨脹或影像侵蝕之其中一種的影像處理；及(A-3)藉由該伺服端，對屬於該原生印鑑影像的每一印鑑影像，及每一進行過影像膨脹或影像侵蝕之其中一種影像處理的印鑑影像，進行色彩變換處理，以獲得與處理前之該印鑑影像具相異顏色的該異色印鑑影像。 According to the method for establishing a seal identification model according to claim 1, this step (A) also includes the following sub-steps: (A-1) For each seal image, use the server to determine whether the seal image is The simulated seal image; (A-2) when the server determines that the seal image is the simulated seal image, the server performs image processing of one of image expansion or image erosion on the seal image; and (A-3) Through the server, each seal image belonging to the native seal image, and each seal image that has undergone one of image expansion or image erosion, is subjected to color conversion processing to obtain and The different-color seal image with the different color of the seal image before processing. 如請求項1所述的印鑑辨識模型建立方法，其中，該步驟(E)包含以下子步驟：(E-1)藉由該伺服端，將重複步驟(B)、(C)與(D)所產生的所有訓練資料分為一訓練子集及一測試子集；(E-2)藉由該伺服端，根據該訓練子集中的每一訓練資料中的該用印影像及該標記位置資料，利用一物件偵測演算法建立用於辨識出一待辨識影像中對應於一戳印物件之物件區域的一訓練模型；(E-3)藉由該伺服端，根據該測試子集中的每一訓練 資料中的該用印影像及該標記位置資料，判斷出該訓練模型的辨識正確率是否高於一門檻值；(E-4)當該伺服端判斷出該訓練模型的辨識正確率並未高於該門檻值時，藉由該伺服端，調整該訓練模型，並進行該子步驟(E-3)；及(E-5)當該伺服端判斷出該訓練模型的辨識正確率高於該門檻值時，該伺服端確認該訓練模型為該印鑑辨識模型。 The method for establishing a seal identification model according to claim 1, wherein the step (E) includes the following sub-steps: (E-1) through the server, steps (B), (C) and (D) will be repeated All the training data generated are divided into a training subset and a test subset; (E-2) With the server, according to the print image and the mark position data in each training data in the training subset, An object detection algorithm is used to establish a training model for identifying the object area corresponding to a stamped object in a to-be-recognized image; (E-3) With the server, according to each test subset train From the printed image and the mark position data in the data, it is determined whether the recognition accuracy of the training model is higher than a threshold; (E-4) When the server determines that the recognition accuracy of the training model is not higher than When the threshold is reached, the server adjusts the training model and performs the sub-step (E-3); and (E-5) when the server determines that the recognition accuracy of the training model is higher than the threshold Value, the server confirms that the training model is the seal identification model. 一種用於建立印鑑辨識模型的伺服端，該伺服端包含：一伺服端儲存模組，儲存有多張相關於多個戳印的印鑑影像，以及多張相關於多張表單的表單影像，其中，該伺服端儲存模組還儲存有多筆姓名及一用於根據一待處理姓名產生一對應該待處理姓名的模擬戳印圖像的印鑑影像產生器，其中，該伺服端儲存模組所儲存的每一印鑑影像屬於一原生印鑑影像及一模擬印鑑影像之其中一者，屬於該原生印鑑影像的印鑑影像係該伺服端根據一相關於一***的戳印圖像所產生的，屬於該模擬印鑑影像的印鑑影像係該伺服端根據該等姓名中的一目標姓名，利用該印鑑影像產生器產生之對應該目標姓名的模擬戳印圖像所對應產生的；及一伺服端處理模組，電連接該伺服端儲存模組；其中，該伺服端處理模組對該等印鑑影像之每一者，將該印鑑影像進行色彩變換處理，以獲得一與處理前之該印鑑影像具相異顏色的異色印鑑影像，並自該等表單影像 及該等異色印鑑影像中分別選取一目標表單影像及一目標印鑑影像，並將該目標印鑑影像疊合於該目標表單影像之上以產生一用印影像，且根據該用印影像產生一指示出該用印影像中該目標印鑑影像所對應之位置的標記位置資料，並產生一包括該用印影像及對應該用印影像之該標記位置資料的訓練資料，以及根據重複前述過程所產生的所有訓練資料，利用一物件偵測演算法建立一用於辨識出一待辨識影像中對應於一戳印物件之物件區域的印鑑辨識模型。 A server for establishing a seal recognition model, the server includes: a server storage module, storing a plurality of seal images related to a plurality of stamps, and a plurality of form images related to a plurality of forms, wherein The server-side storage module also stores multiple names and a seal image generator for generating a simulated stamp image corresponding to the name to be processed according to a name to be processed, wherein the server-side storage module stores Each stored seal image belongs to one of a native seal image and a simulated seal image. The seal image belonging to the native seal image is generated by the server based on a stamp image related to a seal and belongs to the The imitation imprint image of the imitation imprint image is generated by the server based on a target name among the names and correspondingly generated by the imitation imprint image corresponding to the target name generated by the imprint image generator; and a server-side processing module , Is electrically connected to the server-side storage module; wherein, the server-side processing module performs color conversion processing on each of the seal images to obtain an image that is different from the seal image before processing The different-color seal image of the color, and from the image of the form And a target form image and a target seal image are selected from the different-color seal images, and the target seal image is superimposed on the target form image to generate a print image, and an indication is generated based on the print image Use the mark position data of the position corresponding to the target seal image in the imprinted image, and generate a training data including the imprinted image and the mark position data corresponding to the imprinted image, and use all the training data generated by repeating the foregoing process. An object detection algorithm establishes a seal recognition model for recognizing an object area corresponding to a stamped object in a to-be-recognized image. 如請求項4所述的伺服端，其中，對於每一印鑑影像，藉由該伺服端處理模組，判斷出該印鑑影像是否為該模擬印鑑影像，當該伺服端處理模組判斷出該印鑑影像為該模擬印鑑影像時，對該印鑑影像進行影像膨脹或影像侵蝕之其中一種的影像處理，以及對屬於該原生印鑑影像的每一印鑑影像，及每一進行過影像膨脹或影像侵蝕之其中一種影像處理的印鑑影像，進行色彩變換處理，以獲得與處理前之該印鑑影像具相異顏色的該異色印鑑影像。 The server according to claim 4, wherein, for each seal image, the server processing module determines whether the seal image is the analog seal image, and when the server processing module determines the seal image When the image is the simulated seal image, perform image processing of one of image expansion or image erosion on the seal image, and perform image processing on each seal image belonging to the native seal image, and each of which has undergone image expansion or image erosion An image-processed seal image undergoes color conversion processing to obtain the different-color seal image with a different color from the seal image before processing. 如請求項4所述的伺服端，其中，該伺服端處理模組將所有訓練資料分為一訓練子集及一測試子集，並根據該訓練子集中的每一訓練資料中的該用印影像及該標記位置資料，利用一物件偵測演算法建立用於辨識出一待辨識影像中對應於一戳印物件之物件區域的一訓練模型，且根據該測試子集中的每一訓練資料中的該用印影像及該標記位置資料，判斷出該訓練模型的辨識正確率是否高於一門檻 值，當該伺服端處理模組判斷出該訓練模型的辨識正確率並未高於該門檻值時，該伺服端處理模組調整該訓練模型並重新判斷，當該伺服端處理模組判斷出該訓練模型的辨識正確率高於該門檻值時，該伺服端處理模組確認該訓練模型為該印鑑辨識模型。 The server according to claim 4, wherein the server processing module divides all training data into a training subset and a test subset, and according to the print image in each training data in the training subset And the marked location data, an object detection algorithm is used to create a training model for identifying the object area corresponding to a stamped object in a to-be-recognized image, and based on the training data in each training data in the test subset Using the printed image and the location data of the mark, it is judged whether the recognition accuracy of the training model is higher than a threshold When the server-side processing module determines that the recognition accuracy of the training model is not higher than the threshold value, the server-side processing module adjusts the training model and re-judges. When the server-side processing module determines When the recognition accuracy of the training model is higher than the threshold value, the server-side processing module confirms that the training model is the seal recognition model.
TW109130059A 2020-09-02 2020-09-02 Method for establishing seal identification model and server terminal for establishing seal identification model TWI745068B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
TW109130059A TWI745068B (en) 2020-09-02 2020-09-02 Method for establishing seal identification model and server terminal for establishing seal identification model

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
TW109130059A TWI745068B (en) 2020-09-02 2020-09-02 Method for establishing seal identification model and server terminal for establishing seal identification model

Publications (2)

Publication Number Publication Date
TWI745068B true TWI745068B (en) 2021-11-01
TW202211111A TW202211111A (en) 2022-03-16

Family

ID=79907364

Family Applications (1)

Application Number Title Priority Date Filing Date
TW109130059A TWI745068B (en) 2020-09-02 2020-09-02 Method for establishing seal identification model and server terminal for establishing seal identification model

Country Status (1)

Country Link
TW (1) TWI745068B (en)

Citations (6)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN103208004A (en) * 2013-03-15 2013-07-17 北京英迈杰科技有限公司 Automatic recognition and extraction method and device for bill information area
TWI402769B (en) * 2004-12-28 2013-07-21 Masahiro Noda Variable imprint seal system, seal verification system, seal verification method and seal verification program
CN104112128A (en) * 2014-06-19 2014-10-22 中国工商银行股份有限公司 Digital image processing system applied to bill image character recognition and method
TWI618013B (en) * 2016-12-29 2018-03-11 臺灣中小企業銀行股份有限公司 Automatic deposit cheque system and method of perfomring the same
CN110619642A (en) * 2019-09-05 2019-12-27 四川大学 Method for separating seal and background characters in bill image
TWM609320U (en) * 2020-09-02 2021-03-21 中國信託商業銀行股份有限公司 Server used for creating seal recognition model

Patent Citations (6)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
TWI402769B (en) * 2004-12-28 2013-07-21 Masahiro Noda Variable imprint seal system, seal verification system, seal verification method and seal verification program
CN103208004A (en) * 2013-03-15 2013-07-17 北京英迈杰科技有限公司 Automatic recognition and extraction method and device for bill information area
CN104112128A (en) * 2014-06-19 2014-10-22 中国工商银行股份有限公司 Digital image processing system applied to bill image character recognition and method
TWI618013B (en) * 2016-12-29 2018-03-11 臺灣中小企業銀行股份有限公司 Automatic deposit cheque system and method of perfomring the same
CN110619642A (en) * 2019-09-05 2019-12-27 四川大学 Method for separating seal and background characters in bill image
TWM609320U (en) * 2020-09-02 2021-03-21 中國信託商業銀行股份有限公司 Server used for creating seal recognition model

Also Published As

Publication number Publication date
TW202211111A (en) 2022-03-16

Similar Documents

Publication Publication Date Title
TWI716923B (en) Car damage image generation method and device based on GAN network
JP5889727B2 (en) Electronic symbol marking apparatus and method thereof
CN111742214A (en) Substrate inspection apparatus and method for determining defect type of screen printer
CN110889457B (en) Sample image classification training method and device, computer equipment and storage medium
CN108304243B (en) Interface generation method and device, computer equipment and storage medium
CN110675940A (en) Pathological image labeling method and device, computer equipment and storage medium
CN110728687B (en) File image segmentation method and device, computer equipment and storage medium
CN112598627A (en) Method, system, electronic device and medium for detecting image defects
CN101567063A (en) Information management system, form definition management server and information management method
CN111814716A (en) Seal removing method, computer device and readable storage medium
CN104899632B (en) A kind of combined type two-dimension code label and its generation method and recognition methods
CN114418899A (en) Self-adaptive repairing method and system for self-color printing and readable storage medium
CN111401363A (en) Frame number image generation method and device, computer equipment and storage medium
JP2021026191A (en) Learned machine learning model, image data generation apparatus, and method
TWM609320U (en) Server used for creating seal recognition model
CN112258507A (en) Target object detection method and device of internet data center and electronic equipment
JP2006094040A (en) Profile generating device, program, and profile generating method
TWI745068B (en) Method for establishing seal identification model and server terminal for establishing seal identification model
US8594384B2 (en) Method and system for measuring text-rendering quality
CN111241974B (en) Bill information acquisition method, device, computer equipment and storage medium
CN117115823A (en) Tamper identification method and device, computer equipment and storage medium
CN110956174A (en) Device number identification method
CN116091470A (en) Intelligent industrial control method and device based on 5G technology, electronic equipment and storage medium
CN115620329A (en) Stamp deviation intelligent identification method based on artificial intelligence
CN113177543A (en) Certificate identification method, device, equipment and storage medium