TW202228066A - Image object classification method, system, computer program product, and computer readable medium - Google Patents
Image object classification method, system, computer program product, and computer readable medium Download PDFInfo
- Publication number
- TW202228066A TW202228066A TW110100510A TW110100510A TW202228066A TW 202228066 A TW202228066 A TW 202228066A TW 110100510 A TW110100510 A TW 110100510A TW 110100510 A TW110100510 A TW 110100510A TW 202228066 A TW202228066 A TW 202228066A
- Authority
- TW
- Taiwan
- Prior art keywords
- results
- classification
- image object
- result
- image
- Prior art date
Links
Images
Landscapes
- Image Analysis (AREA)
- Information Retrieval, Db Structures And Fs Structures Therefor (AREA)
Abstract
Description
本發明係關於一種分類技術，特別是關於一種針對圖像物件的多樣性進行多元分類的圖像物件分類方法、系統、電腦程式產品及電腦可讀取紀錄媒體。The present invention relates to a classification technology, in particular to an image object classification method, system, computer program product and computer-readable recording medium for performing multi-classification for the diversity of image objects.
機器學習的開發與應用逐漸成為顯學，以便利用大量數據(或資料)訓練機器學習模型，經由訓練完成的模型來獲得某些預測資訊。The development and application of machine learning has gradually become obvious, so that a large amount of data (or data) can be used to train a machine learning model, and certain prediction information can be obtained through the trained model.
實際上，物件分類技術應用廣泛，例如應用於圖像物件分類等，習知物件分類模型通常強調的是，可將物件歸屬於單一類別的資訊作為輸出結果。In fact, the object classification technology is widely used, for example, it is applied to image object classification, etc. The conventional object classification model usually emphasizes that the information that the object belongs to a single category can be used as the output result.
惟，將物件分成單一類別仍有不足之處，以含有多種圖像類別的數位文件為例，實際應用情境需要將一個物件同時歸屬於多種類別，然此為習知分類通常不予考慮的因素。以往雖有一些分類技術，但仍不適於特定應用。However, classifying objects into a single category still has shortcomings. Taking digital files containing multiple image categories as an example, the actual application situation requires that an object belong to multiple categories at the same time, but this is a factor that is usually not considered in conventional classification. . Although some classification techniques exist in the past, they are still not suitable for specific applications.
有鑑於此，有必要提供一種有別以往的技術方案，以解決習知技術所存在的問題。In view of this, it is necessary to provide a different technical solution to solve the problems existing in the prior art.
本發明之一目的在於提供一種圖像物件分類方法，能夠將圖像物件歸屬於多種類別，有利適用於含有多種圖像類別的數位檔案。One object of the present invention is to provide a method for classifying image objects, which can classify image objects into various categories, and is advantageously applicable to digital files containing various image categories.
本發明之次一目的在於提供一種圖像物件分類系統，能夠將圖像物件歸屬於多種類別，有利適用於含有多種圖像類別的數位檔案。Another object of the present invention is to provide an image object classification system, which can classify image objects into various categories, which is beneficial to be suitable for digital files containing various image categories.
本發明之另一目的在於提供一種電腦程式產品，能夠將圖像物件歸屬於多種類別，有利適用於含有多種圖像類別的數位檔案。Another object of the present invention is to provide a computer program product capable of assigning image objects to multiple categories, which is advantageously applicable to digital files containing multiple image categories.
本發明之再一目的在於提供一種電腦可讀取紀錄媒體，能夠將圖像物件歸屬於多種類別，有利適用於含有多種圖像類別的數位檔案。Another object of the present invention is to provide a computer-readable recording medium capable of assigning image objects to various categories, which is advantageously suitable for digital files containing various image categories.
為達上述之目的，本發明的一方面提供一種圖像物件分類方法，由耦接一記憶體的一處理器執行，包括步驟：提供一圖檔，該圖檔包括至少一圖像物件；對該圖像物件進行數種二元分類化特徵抽取處理，以取得在類別上互相獨立的數個第一結果；將該數個第一結果以基於串接的降維方式進行結合，並對被結合之該數個第一結果以全連接方式進行特徵抽象化處理，以取得一第二結果；及對該數個第一結果及該第二結果以矩陣點乘方式進行特徵整合處理，以取得一分類結果。To achieve the above object, one aspect of the present invention provides an image object classification method, which is executed by a processor coupled to a memory, and includes the steps of: providing an image file, the image file including at least one image object; The image object is subjected to several binary classification feature extraction processes to obtain several first results that are independent of each other in category; the several first results are combined in a dimensionality reduction method based on concatenation, and the Perform feature abstraction processing on the combined first results in a fully connected manner to obtain a second result; and perform feature integration processing on the first results and the second results in a matrix dot product manner to obtain A classification result.
在本發明之一實施例中，該分類結果包含該第二結果，該第二結果還包含該數個第一結果的分類信任度，該圖像物件分類方法在取得該分類結果後，還包括步驟：對該分類結果進行文檔化處理，依據該數個第一結果的分類信任度進行排序的結果挑選該數個第一結果中的至少一個，將該被挑選的至少一個第一結果對應的至少一類別屬性名稱、至少一物件位置及至少一物件尺寸記錄於一文檔。In an embodiment of the present invention, the classification result includes the second result, and the second result further includes classification confidence levels of the plurality of first results. After obtaining the classification result, the image object classification method further includes: Step: documenting the classification results, selecting at least one of the first results according to the results of sorting the classification trust degrees of the first results, and selecting the corresponding at least one first result. At least one category attribute name, at least one object position and at least one object size are recorded in a file.
在本發明之一實施例中，該物件位置包括以下組合中的任一組合：該圖像物件的一起點座標及一終點座標的一組合；或該圖像物件的一中心座標、一物件長度及一物件寬度的一組合。In an embodiment of the present invention, the object position includes any combination of the following combinations: a combination of a start point coordinate and an end point coordinate of the image object; or a center coordinate and an object length of the image object and a combination of object widths.
在本發明之一實施例中，該圖像物件分類方法在取得該分類結果後，還包括步驟：對該分類結果進行圖文化處理，使該分類結果以圖塊、字塊或其組合呈現在該圖檔中。In an embodiment of the present invention, after obtaining the classification result, the image object classification method further includes the step of: performing graphic culture processing on the classification result, so that the classification result is presented in the image block, word block, or a combination thereof. in this file.
在本發明之一實施例中，取得該第二結果的步驟包括：對被結合之該數個第一結果以多層感知方式進行特徵抽象化處理，以取得該第二結果。In an embodiment of the present invention, the step of obtaining the second result includes: performing feature abstraction processing on the combined first results in a multi-layer perceptual manner to obtain the second result.
在本發明之一實施例中，將該數個第一結果以基於串接的降維方式進行結合的步驟包括：將該數個第一結果依序進行串接，以使該數個第一結果被結合而形成一合成降維結果。In an embodiment of the present invention, the step of combining the plurality of first results in a concatenation-based dimensionality reduction manner includes: sequentially concatenating the plurality of first results, so that the plurality of first results are The results are combined to form a composite dimensionality reduction result.
在本發明之一實施例中，對該圖像物件進行數種二元分類化特徵抽取處理的步驟包括：將該圖像物件以數種圖像類別進行特徵抽取處理，該數種圖像類別包括一電子零件的數種特徵示意圖。In an embodiment of the present invention, the step of performing several binary classification feature extraction processing on the image object includes: performing feature extraction processing on the image object with several image categories, the several image categories Includes several feature schematics of an electronic component.
為達上述之目的，本發明的另一方面提供一種圖像物件分類系統，包括一處理器及一記憶體，該處理器耦接該記憶體，該記憶體儲存至少一指令，該處理器執行該指令，以執行如上所述之圖像物件分類方法。To achieve the above object, another aspect of the present invention provides an image object classification system, comprising a processor and a memory, the processor is coupled to the memory, the memory stores at least one instruction, and the processor executes This command is used to execute the image object classification method as described above.
為達上述之目的，本發明的又一方面提供一種電腦程式產品，當電腦載入該電腦程式並執行後，該電腦能夠執行如上所述之圖像物件分類方法。In order to achieve the above-mentioned object, another aspect of the present invention provides a computer program product. After the computer program is loaded and executed, the computer can execute the above-mentioned image object classification method.
為達上述之目的，本發明的再一方面提供一種電腦可讀取紀錄媒體，該電腦可讀取紀錄媒體內儲程式，當電腦載入該程式並執行後，該電腦能夠完成如上所述之圖像物件分類方法。In order to achieve the above-mentioned purpose, another aspect of the present invention provides a computer-readable recording medium, the computer can read a program stored in the recording medium, and after the computer loads the program and executes it, the computer can complete the above-mentioned Image object classification method.
本發明的圖像物件分類方法、系統、電腦程式產品及電腦可讀取紀錄媒體，通過提供一圖檔，該圖檔包括至少一圖像物件；對該圖像物件進行數種二元分類化特徵抽取處理，以取得在類別上互相獨立的數個第一結果；將該數個第一結果以基於串接的降維方式進行結合，並對被結合之該數個第一結果以全連接方式進行特徵抽象化處理，以取得一第二結果；及對該數個第一結果及該第二結果以矩陣點乘方式進行特徵整合處理，以取得一分類結果。藉由上述物件分類過程，能夠輸出圖像物件歸屬於多種類別的隱含資訊，有利適用於含有多種圖像類別的數位文件。The image object classification method, system, computer program product, and computer-readable recording medium of the present invention provide an image file including at least one image object; perform several binary classifications on the image object Feature extraction processing to obtain several first results that are independent of each other in category; combine the several first results in a dimensionality reduction method based on concatenation, and combine the several first results with a full connection performing feature abstraction processing in a manner to obtain a second result; and performing feature integration processing on the plurality of first results and the second results in a matrix dot product manner to acquire a classification result. Through the above-mentioned object classification process, the implicit information that the image objects belong to various categories can be output, which is advantageously applicable to digital files containing various image categories.
為了讓本發明之上述及其他目的、特徵、優點能更明顯易懂，下文將特舉本發明較佳實施例，並配合所附圖式，作詳細說明如下。再者，本發明所提到的方向用語，例如上、下、頂、底、前、後、左、右、內、外、側面、周圍、中央、水平、橫向、垂直、縱向、軸向、徑向、最上層或最下層等，僅是參考附加圖式的方向。因此，使用的方向用語是用以說明及理解本發明，而非用以限制本發明。In order to make the above-mentioned and other objects, features and advantages of the present invention more clearly understood, the preferred embodiments of the present invention will be exemplified below and described in detail in conjunction with the accompanying drawings. Furthermore, the directional terms mentioned in the present invention, such as up, down, top, bottom, front, rear, left, right, inner, outer, side, surrounding, center, horizontal, lateral, vertical, longitudinal, axial, Radial, uppermost or lowermost, etc., are only directions with reference to the attached drawings. Therefore, the directional terms used are for describing and understanding the present invention, not for limiting the present invention.
請參閱第1圖所示，本發明的一方面提供一種圖像物件分類系統，該圖像物件分類系統可被配置成包括一處理裝置1，該處理裝置1包括相互耦接的一分類模組11及一格式轉換模組12，該耦接方式可為有線連接、無線傳輸或資料交換等耦合或連接方式，該分類模型11及該格式轉換模型12可以是軟體模組、硬體模組或軟硬體協同運作模組；例如：該分類模組11可被配置成具有資料輸入、處理及輸出功能，譬如：數據讀取、運算及顯示等功能，用以依據一外部資料D產生一分類結果R；該格式轉換模型12可被配置成具備資料格式轉換功能；此外，該圖像物件分類系統還可包括一資料庫(圖未繪示)，該資料庫可耦接該分類模型11及該格式轉換模型12，用於儲存進行資料處理的相關資料，如訓練機器學習模型用的資料集或參數等；該圖像物件分類系統還可包括其他部分，例如：一零件分類單元A1及一零件視圖辨識單元A2，分別用於依據該分類結果R進行一零件分類作業及一零件視圖方向辨識作業。Referring to FIG. 1, an aspect of the present invention provides an image object classification system, the image object classification system can be configured to include a
示例地，該圖像物件分類系統可被配置成包括：一處理器及一記憶體，該處理器耦接該記憶體，該記憶體儲存至少一指令，該處理器執行該指令，以執行本發明另一方面提供的一種圖像物件分類方法，其係舉例說明於後，惟不以此為限。For example, the image object classification system can be configured to include: a processor and a memory, the processor is coupled to the memory, the memory stores at least one instruction, the processor executes the instruction to execute the present invention Another aspect of the invention provides a method for classifying image objects, which is illustrated in the following, but not limited thereto.
本發明的另一方面提供一種圖像物件分類方法，由耦接一記憶體的一處理器執行，包括：提供一圖檔，譬如該圖檔可由一文件檔轉換而成，該圖檔包括至少一圖像物件；對該圖像物件進行數種二元分類化特徵抽取處理，以取得在類別上互相獨立的數個第一結果；將該數個第一結果以基於串接的降維方式進行結合，並對被結合之該數個第一結果以全連接方式進行特徵抽象化處理，以取得一第二結果；及對該數個第一結果及該第二結果以矩陣點乘方式進行特徵整合處理，以取得一分類結果。以下舉例說明可被實施的示例態樣，作為瞭解相關內容的說明示例，惟不以此為限。Another aspect of the present invention provides an image object classification method, executed by a processor coupled to a memory, including: providing an image file, for example, the image file can be converted from a file file, the image file includes at least an image object; perform several binary classification feature extraction processes on the image object to obtain several first results that are independent of each other in category; use the several first results to reduce the dimension based on concatenation combine and perform feature abstraction processing on the combined first results in a fully connected manner to obtain a second result; and perform matrix dot product on the several first results and the second results Feature integration processing to obtain a classification result. The following examples illustrate example aspects that may be implemented, as illustrative examples for understanding the relevant content, but not limited thereto.
舉例而言，如第2圖所示，該圖像物件分類方法可以包括步驟S1至S6，該些步驟的部分可依實際應用被適當地換序、簡化或省略，以完成上述方法實施例的至少一部分。For example, as shown in FIG. 2, the image object classification method may include steps S1 to S6, and some of these steps may be appropriately reordered, simplified or omitted according to practical applications, so as to complete the above method embodiments. at least part of it.
步驟S1，可輸入外部資料，作為後續進行圖像物件分類的依據，例如：讀取外部資料，該外部資料可為一圖檔，譬如該圖檔可預先由一外部機器將一文件檔進行檔案轉檔及儲存而被提供，但不以此為限，該外部資料也可為該文件檔，可由該處理器對該文件檔進行轉換而提供該圖檔，該圖檔之提供方式至少如上所述，在此不作限制。示例地，該文件檔可以例如是內含各式圖例(譬如不同電子零件視圖)及字例(譬如說明文字)的數位文件檔，譬如「*.pdf」，其係一種可攜式文件格式(Portable Document Format)，例如檔案內容可包含有關電子零件的技術文件(datasheet)；但不以此為限，該數位文件檔也可以是其他檔案格式，譬如「*.doc」或「*.odt」等包括圖文的文件格式；但不以此為限，該數位文件檔還可以是其他檔案格式，譬如「*.fpk」，其為一種富比庫格式(footprintku format)，屬於一種由富比庫(FOOTPRINTKU)公司制定的資料儲存格式，該資料儲存格式可以儲存記錄的資料涵蓋電子零件參數、圖形及文件。此外，該外部資料還可以包括其他資料，譬如表格等；該圖檔可以是經過壓縮或未經壓縮的檔案格式，譬如「*.jpg」，但不以此為限，該圖檔也可以是其他檔案格式，譬如「*.png」或「*.bmp」等，後續所述圖檔可以意指檔案資料或其被呈現在一顯示裝置時的畫面內容，也可被稱為圖片。後續，可進行步驟S2。In step S1, external data can be input as a basis for subsequent classification of image objects, for example: reading external data, the external data can be an image file, for example, the image file can be pre-filed by an external machine as a file provided by converting and storing, but not limited to, the external data can also be the document file, and the processor can convert the document file to provide the image file, and the image file is provided in at least the above-mentioned manner not limited here. For example, the file can be, for example, a digital file containing various legends (such as views of different electronic components) and font examples (such as explanatory text), such as "*.pdf", which is a portable file format ( Portable Document Format), for example, the content of the file can include technical documents related to electronic components (datasheet); but not limited to this, the digital file can also be in other file formats, such as "*.doc" or "*.odt" etc. including graphic file formats; but not limited to this, the digital file can also be in other file formats, such as "*.fpk", which is a Footprintku format, which belongs to a The data storage format developed by the FOOTPRINTKU company can store the recorded data covering electronic component parameters, graphics and documents. In addition, the external data can also include other data, such as tables, etc.; the image file can be in a compressed or uncompressed file format, such as "*.jpg", but not limited to this, the image file can also be For other file formats, such as "*.png" or "*.bmp", the following image file may refer to file data or its screen content when it is displayed on a display device, and may also be referred to as a picture. Subsequently, step S2 may be performed.
步驟S2，可對該至少一圖像物件進行數種二元分類化特徵抽取處理，以取得在類別上互相獨立的數個第一結果。例如：將含有該至少一圖像物件的圖檔同時送入數個具備二元分類器功能的模型，譬如用以取得在類別上的分類型態互相獨立的數個第一結果，其中模型特徵(如架構及參數等)設定方式係其所屬技術領域中具有通常知識者可以理解，不再贅述。後續，可進行步驟S3。In step S2, several binary classification feature extraction processes may be performed on the at least one image object, so as to obtain several first results that are independent of each other in category. For example: the image file containing the at least one image object is simultaneously fed into several models with binary classifier functions, for example, to obtain several first results whose classification states are independent of each other, wherein the model features The setting method (such as the structure and parameters, etc.) can be understood by those with ordinary knowledge in the technical field to which it belongs, and will not be repeated here. Subsequently, step S3 may be performed.
舉例而言，資料集中的圖像資料可被輸入數個已針對不同特徵訓練完成的二元分類模型進行運算，以產生數個輸出結果作為該數個第一結果。For example, the image data in the data set can be input into several binary classification models that have been trained for different features for operation, so as to generate several output results as the several first results.
示例地，該數個具備二元分類器功能的模型可以針對不同圖像物件特徵進行特徵抽取。以電子零件的應用情境為例，可針對電子零件的各種特徵示意圖例進行分類，例如外觀特徵、電性特徵及應用特徵等特徵示意圖，諸如電子零件的方向視圖(譬如上視、下視及側視等三視圖，通常有用於呈現零件尺寸的幾何圖形搭配線段等)、電子零件的腳位圖(譬如積體電路封裝腳位圖，通常有表示接腳的幾何圖形搭配文字等)、電路圖(譬如由電子元件符號連接形成的電路，通常有幾何圖形搭配線條等)、特性曲線(譬如電子零件的電壓或電流等諸多特性曲線，通常有連續延伸的線條構成的波形資訊等)及訊號時序圖(譬如電子零件的時脈、輸入、輸出等訊號的時序圖，通常有連續延伸的線條構成的封閉塊狀波形用於呈現連續波形關係，其與特性曲線主要差異在於有較多封閉區塊及轉折線條等)，惟不以此為限。For example, the several models with the function of binary classifiers can perform feature extraction for different image object features. Taking the application situation of electronic parts as an example, various characteristic schematic examples of electronic parts can be classified, such as appearance features, electrical characteristics and application characteristics, such as the directional views of electronic parts (such as top view, bottom view and side view). three views, usually including geometric figures used to present the size of the parts with line segments, etc.), pin diagrams of electronic parts (such as integrated circuit package pin diagrams, usually with geometric figures representing pins with text, etc.), circuit diagrams ( For example, the circuit formed by the symbolic connection of electronic components, usually has geometric figures with lines, etc.), characteristic curves (such as the voltage or current of electronic parts and many other characteristic curves, usually waveform information formed by continuously extending lines, etc.) and signal timing diagrams (For example, the timing diagram of the clock, input, output and other signals of electronic parts, usually there is a closed block waveform composed of continuously extending lines to present a continuous waveform relationship. The main difference from the characteristic curve is that there are more closed blocks and turning lines, etc.), but not limited to this.
舉例而言，在其他圖像物件分類應用情境中，譬如手機使用說明書，則可能包括外觀功能示意圖及螢幕畫面功能示意圖等，也可針對其圖例特徵加以分析，作為圖像物件分類的依據。For example, in other application scenarios of image object classification, such as a mobile phone user manual, it may include a schematic diagram of appearance functions and a schematic diagram of screen functions.
步驟S3，可將該數個第一結果以基於串接的降維方式(即將該數個第一結果的數據依序串接，使在平面上排列之數據降低維度，轉成線性排列之數據)進行結合，並對被結合之該數個第一結果以全連接方式進行特徵抽象化處理，以取得一第二結果。例如：將該數個第一結果進行一合成降維處理(如二維資料轉成一維資料)，譬如該合成降維處理是將該數個第一結果依序進行串接，以形成一合成降維結果，再將該合成降維結果進行特徵抽象化處理，例如依照該合成降維結果的特徵(模式)歸納為資訊量更少的重要表徵，忽略無關緊要的細節，降低複雜度，變為具備計算能力的裝置容易處理的模型，譬如可使用多層感知機(也可稱為全連接層)，例如可使用1×1卷積(convolution)進行維度轉換，使用空洞卷積(dilated convolution)調整感受野增加或減少抽取的特徵量，以取得該第二結果，例如該第二結果包括該數個第一結果的象徵值，譬如數個第一結果的分類信任度(諸如所有類別的值域都是獨立且介於0~1之間，Sigmoid[0:1])，其設定方式係所屬技術領域者可以理解，不再贅述。後續，可進行步驟S4。In step S3, the plurality of first results can be reduced in dimension based on concatenation (that is, the data of the plurality of first results can be concatenated in sequence to reduce the dimension of the data arranged on the plane and convert it into linearly arranged data. ) are combined, and feature abstraction processing is performed on the combined first results in a fully connected manner to obtain a second result. For example, performing a synthetic dimension reduction process on the first results (such as converting two-dimensional data into one-dimensional data), for example, in the synthetic dimension reduction process, the multiple first results are sequentially concatenated to form a Synthesize the dimensionality reduction results, and then perform feature abstraction processing on the synthetic dimensionality reduction results. For example, according to the features (patterns) of the synthetic dimensionality reduction results, they are summarized into important representations with less information, ignoring insignificant details and reducing complexity. It becomes a model that can be easily processed by a device with computing power, for example, a multi-layer perceptron (also called a fully connected layer) can be used, for example, a 1×1 convolution can be used for dimension transformation, and a dilated convolution can be used. ) adjust the receptive field to increase or decrease the amount of extracted features to obtain the second result, for example, the second result includes the symbolic values of the first results, such as the classification confidence of the first results (such as all categories of The value ranges are all independent and between 0 and 1, Sigmoid[0:1]), and the setting method thereof can be understood by those skilled in the art, and will not be repeated here. Subsequently, step S4 may be performed.
步驟S4，可對該數個第一結果及該第二結果以矩陣點乘方式進行特徵整合處理，以取得一分類結果。例如將該第二結果送入一個具備特徵分類器功能的模型，例如具備全連接層的神經網路模型，本發明實施例在此所採用的全連接層與另一種使用1×1卷積的全連接層在整體應用情境上有所不同。例如：如第3圖所示，在本例中，經由二元分類器所取得的數個第一結果除了提供在此所採用的全連接層作為輸入資料之外，該數個第一結果同時亦將提供給一特徵整合器作為運算來源資訊，例如全連接層輸出的該第二結果會與該數個第一結果進行一整合運算，譬如可為疊加運算，例如可採用矩陣點乘運算，譬如將在兩矩陣依序取相應位置的元素進行乘法運算所取得的乘積當作另一矩陣在相應位置的元素，使該特徵整合器的運算結果包含諸多原始的二元分類資訊及由其衍生的抽取分類資訊，以整合多種分類資訊作為該分類結果，其模型特徵(如架構及參數等)設定方式係其所屬技術領域中具有通常知識者可以理解，不再贅述於此。後續，可進行步驟S5及/或步驟S6。In step S4, feature integration processing may be performed on the plurality of first results and the second results by means of matrix dot product, so as to obtain a classification result. For example, the second result is sent to a model with a feature classifier function, such as a neural network model with a fully connected layer. The fully connected layer is different in the overall application context. For example: as shown in Figure 3, in this example, in addition to providing the fully connected layer used here as the input data, the first results obtained through the binary classifier are simultaneously It will also be provided to a feature integrator as operation source information. For example, the second result output by the fully connected layer will perform an integration operation with the first results, such as a superposition operation, such as a matrix dot product operation, For example, the product obtained by multiplying the elements in the corresponding positions of the two matrices in sequence is regarded as the elements of the other matrix in the corresponding positions, so that the operation result of the feature integrator contains a lot of original binary classification information and its derivatives. The extracted classification information is used to integrate various classification information as the classification result. The setting method of the model features (such as structure and parameters, etc.) can be understood by those with ordinary knowledge in the technical field to which they belong, and will not be repeated here. Subsequently, step S5 and/or step S6 may be performed.
可選地，步驟S5，該分類結果包含該第二結果，該第二結果還包含該數個第一結果的分類信任度，因此，還可在取得該分類結果後，對該分類結果進行文檔化處理，依據該數個第一結果的排序結果從中挑選至少一第一結果記錄至少一類別屬性名稱、至少一物件位置及至少一物件尺寸，譬如在取得該分類結果時，由於該分類結果包含該第二結果中的資訊，且該第二結果包含數個第一結果的分類信任度，因此，可將該些第一結果依分類信任度進行排序，譬如將該些第一結果的分類信任度進行正規化，例如正規化方式可以為Max-Min Normalization, L1 Normalization, L2 Normalization等，其實施方式係其所屬技術領域中具有通常知識者可以理解，不再贅述；後續，可將該些正規化的信任度進行排序，依照該排序結果分別記錄該些第一結果各自對應的類別屬性名稱、物件位置及物件尺寸；後續，可依據預設的條件挑選該數個信任度中的至少一個第一結果對應的類別屬性名稱、物件位置及物件尺寸建立一文檔，例如挑選信任度最大者或前幾大者，以便節省該文檔的資料儲存空間及資料傳輸量。例如：該物件位置包括以下內容中的任一種：該圖像物件的一起點座標(譬如左上角座標)及一終點座標(譬如右下角座標)；或者，該圖像物件的一中心座標、一物件長度及一物件寬度，惟不以此為限。Optionally, in step S5, the classification result includes the second result, and the second result also includes the classification confidence of the first results. Therefore, after obtaining the classification result, the classification result can be documented. processing, selecting at least one first result according to the sorting results of the plurality of first results to record at least one category attribute name, at least one object position and at least one object size, for example, when the sorting result is obtained, because the sorting result contains The information in the second result, and the second result includes the classification confidence of several first results, therefore, the first results can be sorted according to the classification confidence, for example, the classification confidence of the first results For example, the normalization methods can be Max-Min Normalization, L1 Normalization, L2 Normalization, etc. The implementation method is understandable to those with ordinary knowledge in the technical field to which they belong, and will not be repeated; in the following, these normalization methods can be used. Sort by the highest trust degree, and record the corresponding category attribute name, object position and object size of the first results according to the sorting result; subsequently, at least one of the several trust degrees can be selected according to the preset conditions. Create a document with the category attribute name, object location and object size corresponding to a result, for example, select the one with the highest trust degree or the top ones, so as to save the data storage space and data transmission volume of the document. For example: the object position includes any one of the following contents: the coordinates of a starting point of the image object (such as the coordinates of the upper left corner) and an end coordinate (such as the coordinates of the lower right corner); or, a center coordinate, a coordinate of the image object The length of an object and the width of an object, but not limited thereto.
可選地，步驟S6，可在取得該分類結果後，將該分類結果進行圖文化處理，使該分類結果以圖塊、字塊或其組合呈現在該圖檔中，例如：將該分類結果以圖像(譬如不同顏色的圖框等)搭配文字(譬如帶有不同底色的黑色文字等)的方式呈現，譬如可在圖塊最上方貼上色塊，並於色塊中以文字填寫分類結果的類別，當該圖塊出現多類別情形時，便於以不同顏色塊進行區分類別差異，避免造成混淆；但不以此為限，例如另一作法可將該圖檔及該分類結果分別儲存在以不同類別名稱命名的資料夾中，使得同一圖檔可同時出現於多個資料夾中，惟不以此為限。Optionally, in step S6, after the classification result is obtained, the classification result can be subjected to graphic culture processing, so that the classification result is presented in the graphic file as a block, a word block or a combination thereof, for example: the classification result It is presented in the form of images (such as picture frames of different colors, etc.) and text (such as black text with different background colors, etc.), for example, a color block can be pasted on the top of the block and filled in with text in the color block The category of the classification result. When the image block has multiple categories, it is convenient to use different color blocks to distinguish the difference between the categories to avoid confusion; but not limited to this, for example, another method can be used to separate the image file and the classification result. Stored in folders named with different category names, so that the same image file can appear in multiple folders at the same time, but not limited to this.
可選地，在一實施例中，該分類結果包含該第二結果，該第二結果還包含該數個第一結果的分類信任度，該圖像物件分類方法在取得該分類結果後，還包括步驟：將該分類結果進行文檔化處理，依據該數個第一結果的分類信任度進行排序的結果挑選該數個第一結果中的至少一個，將該被挑選的至少一個第一結果對應的至少一類別屬性名稱、至少一物件位置及至少一物件尺寸紀錄於一文檔。藉此，可將分類結果進行文檔化，使該分類結果被儲存為用戶易於理解的欄位內容，有助於進行相關分析。Optionally, in one embodiment, the classification result includes the second result, and the second result also includes the classification confidence of the plurality of first results, and the image object classification method further includes the classification result after obtaining the classification result. It includes the steps of: documenting the classification results, selecting at least one of the first results according to the results of sorting the classification trust degrees of the first results, and corresponding to the selected at least one first result The at least one category attribute name, the at least one object position and the at least one object size are recorded in a document. Thereby, the classification result can be documented, so that the classification result can be stored as the content of the field that is easy for the user to understand, which is helpful for related analysis.
可選地，在一實施例中，該物件位置包括以下組合中的任一組合：該圖像物件的一起點座標及一終點座標的一組合；或該圖像物件的一中心座標、一物件長度及一物件寬度的一組合。藉此，可將物件位置的格式設定為以座標為基礎的參數，有利於後續進行影像處理(例如貼圖等)。Optionally, in one embodiment, the object position includes any combination of the following combinations: a combination of a start point coordinate and an end point coordinate of the image object; or a center coordinate of the image object, an object A combination of length and width of an object. In this way, the format of the object position can be set as a coordinate-based parameter, which is beneficial for subsequent image processing (such as mapping, etc.).
可選地，在一實施例中，該圖像物件分類方法在取得該分類結果後，還包括步驟：對該分類結果進行圖文化處理，使該分類結果以圖塊、字塊或其組合呈現在該圖檔中。藉此，可將分類結果進行圖文化，使該分類結果被直接呈現在該圖檔中，有助於用戶直觀地得知該分類結果。Optionally, in an embodiment, after obtaining the classification result, the image object classification method further includes the step of: performing a graphic culture process on the classification result, so that the classification result is presented in image blocks, word blocks or a combination thereof in this file. In this way, the classification result can be image cultured, so that the classification result can be directly presented in the image file, which is helpful for the user to know the classification result intuitively.
可選地，在一實施例中，取得該第二結果的步驟包括：對被結合之該數個第一結果以多層感知方式進行特徵抽象化處理，以取得該第二結果。藉此，通過該多層感知方式將該數個第一結果的特徵抽象化，有利於簡化中間資料處理複雜度。Optionally, in an embodiment, the step of obtaining the second result includes: performing feature abstraction processing on the combined first results in a multi-layer perceptual manner to obtain the second result. Thereby, the features of the first results are abstracted through the multi-layer perceptual method, which is beneficial to simplify the processing complexity of the intermediate data.
可選地，在一實施例中，將該數個第一結果以基於串接的降維方式進行結合的步驟包括：將該數個第一結果依序進行串接，以使被結合的該數個第一結果形成一合成降維結果。藉此，可彙整該數個第一結果包含的所有資訊，以確保資料的豐富性，而且，還可簡化該合成降維處理所需的運算量，有利於縮短該分類結果的取得時程。Optionally, in an embodiment, the step of combining the plurality of first results in a concatenation-based dimensionality reduction manner includes: sequentially concatenating the plurality of first results, so that the combined first results are Several first results form a composite dimensionality reduction result. In this way, all the information contained in the first results can be integrated to ensure the richness of the data, and the computation amount required for the synthetic dimension reduction processing can also be simplified, which is beneficial to shorten the time for obtaining the classification results.
可選地，在一實施例中，對該圖像物件進行數種二元分類化特徵抽取處理的步驟包括：將該圖像物件以數種圖像類別進行特徵抽取處理，該數種圖像類別包括一電子零件的數種特徵示意圖，諸如電子零件的方向視圖、電子零件的腳位圖、電路圖、特性曲線及訊號時序圖等，惟不以此為限。藉此，可以針對不同圖像類別，諸如電子零件的方向視圖、電子零件的腳位圖、電路圖、特性曲線及訊號時序圖，可被有效分類，有利於相關人員加速進行資料判讀、分析及相關開發時程。Optionally, in one embodiment, the step of performing several binary classification feature extraction processing on the image object includes: performing feature extraction processing on the image object in several image categories, the The category includes several characteristic diagrams of an electronic component, such as the orientation view of the electronic component, the pin diagram of the electronic component, the circuit diagram, the characteristic curve and the signal timing diagram, etc., but not limited thereto. In this way, different image categories, such as the orientation view of electronic parts, the pin map of electronic parts, circuit diagrams, characteristic curves and signal timing diagrams, can be effectively classified, which is helpful for relevant personnel to accelerate data interpretation, analysis and correlation. development schedule.
另一方面，本發明還提供一種圖像物件分類系統，包括一處理器及一記憶體，該處理器耦接該記憶體，其耦接方式可為有線或無線方式，該記憶體儲存至少一指令，該處理器執行該指令，以執行如上所述之圖像物件分類方法。In another aspect, the present invention also provides an image object classification system, comprising a processor and a memory, the processor is coupled to the memory, and the coupling mode can be wired or wireless, and the memory stores at least one instruction, the processor executes the instruction to perform the image object classification method as described above.
舉例而言，該圖像物件分類系統可被配置成具有資料處理功能的電子裝置，例如：雲端平台機器、伺服器、桌上型電腦、筆記型電腦、平板電腦或智慧型手機等，惟不以此為限，用於執行如上所述之圖像物件分類方法。其實施方式已說明如上，不再贅述。For example, the image object classification system can be configured as an electronic device with data processing functions, such as a cloud platform machine, a server, a desktop computer, a notebook computer, a tablet computer or a smart phone, etc., but not As a limitation, it is used to implement the image object classification method as described above. Its implementation has been described above, and will not be repeated here.
另一方面，本發明還提供一種電腦程式產品，當電腦載入該電腦程式並執行後，該電腦能夠執行如上所述之圖像物件分類方法。例如：該電腦程式產品可包含數個程式指令，該程式指令可利用現有的程式語言實現，以便用於執行如上所述之圖像物件分類方法，例如：以Python搭配Numpy、Matplotlib及Tensorflow套件等，惟不以此為限。On the other hand, the present invention also provides a computer program product. After the computer program is loaded and executed, the computer can execute the above-mentioned image object classification method. For example, the computer program product may include several program instructions, which may be implemented using existing programming languages for implementing the above-described image object classification method, such as: Python with Numpy, Matplotlib, and Tensorflow packages, etc. , but not limited to this.
另一方面，本發明還提供一種電腦可讀取紀錄媒體，例如：光碟、隨身碟或硬碟等，但不以此為限，應被理解的是，電腦可讀取紀錄媒體也可被配置成其他形式的電腦資料儲存媒體，譬如雲端儲存空間(諸如One Drive、Google Drive、Azue Blob或其組合等)或資料伺服器或虛擬機器等，該電腦可讀取紀錄媒體內儲程式(如上述電腦程式)，當電腦載入該程式並執行後，該電腦能夠完成如上所述之圖像物件分類方法。On the other hand, the present invention also provides a computer-readable recording medium, such as an optical disc, a flash drive or a hard disk, etc., but not limited to this. It should be understood that the computer-readable recording medium can also be configured with into other forms of computer data storage media, such as cloud storage space (such as One Drive, Google Drive, Azue Blob or a combination thereof, etc.) or data servers or virtual machines, etc. computer program), when the computer loads and executes the program, the computer can complete the image object classification method as described above.
為了使相關人員能更為理解本發明的實施例特點，以下係以輸入有關電子零件的技術文件(datasheet)作為文件檔為例，示例性以圖解說明上述實施例物件分類過程，但並非旨在以此作為限制。In order to enable the relevant persons to better understand the features of the embodiments of the present invention, the following is an example of inputting a technical file (datasheet) related to electronic components as a file file, which is exemplified to illustrate the object classification process of the above-mentioned embodiment, but it is not intended to Use this as a limit.
值得注意的是，本發明上述實施例的圖像物件分類方法、系統、電腦程式產品及電腦可讀取紀錄媒體，主要採用一多重分類器X(如第3圖所示)的概念，先將輸入的圖像物件進行數種二元分類化(Binary Multi-classification)特徵抽取處理，以取得在類別上互相獨立的數個第一結果，例如可將該數個第一結果以基於串接的降維方式進行結合，並對被結合之該數個第一結果以全連接方式進行特徵抽象化，以取得一第二結果，再對該數個第一結果及該第二結果以矩陣點乘方式進行特徵整合處理，以取得輸出的分類結果(例如所有類別的值域都是獨立且介於0~1之間，Sigmoid[0:1])。It is worth noting that the image object classification method, system, computer program product, and computer-readable recording medium of the above-mentioned embodiments of the present invention mainly adopt the concept of a multi-classifier X (as shown in FIG. 3 ). Perform several binary multi-classification (Binary Multi-classification) feature extraction processes on the input image object to obtain several first results that are independent of each other in category. For example, the several first results can be based on concatenation Combine the dimensionality reduction methods of , and perform feature abstraction on the combined first results in a fully connected manner to obtain a second result, and then use the matrix points for the first results and the second results. The feature integration process is performed by multiplication to obtain the output classification result (for example, the value ranges of all categories are independent and between 0 and 1, Sigmoid[0:1]).
相較之下，另一種非多重分類器Y(如第4圖所示)，其雖可用於圖片分類，大多採用以卷積神經網路(Convolutional Neural Network，簡稱CNN)為基底的模型，諸如超解析度測試序列 (Visual Geometry Group，簡稱VGG)、深度殘差(Deep Residual Network，簡稱ResNet)及即時物件偵測模型(You only look once: Unified, Real-Time Object Detection，簡稱Yolo)等模型，將輸入的圖像經過卷積、池化及全連接化，以取得輸出的分類結果(例如所有類別相加的和為1，Softmax[0:1])。但是，此類型的模型進行分類作業後，最終只能得出單一分類結果，與本發明不同。In contrast, another non-multiple classifier Y (as shown in Figure 4), although it can be used for image classification, mostly uses a model based on a Convolutional Neural Network (CNN), such as Super-resolution test sequence (Visual Geometry Group, VGG for short), Deep Residual Network (ResNet for short) and real-time object detection model (You only look once: Unified, Real-Time Object Detection, referred to as Yolo) and other models , the input image is subjected to convolution, pooling and full connection to obtain the output classification result (for example, the sum of all categories is 1, Softmax[0:1]). However, after this type of model performs the classification operation, only a single classification result can be finally obtained, which is different from the present invention.
相較地，本發明上述實施例的圖像物件分類方法，採用複數個二元的分類器，每一個分類器只針對兩個類別進行分類(譬如需要分成5個類別時，可使用10個分類器)，其優點至少在於：本發明上述實施例的圖像物件分類方法可以透過每一個二元分類結果得知所有類別中有哪些容易區分及有哪些不易區分，並在後面加上一個全連接層進行特徵的整合訓練，例如於輸出時可呈現數種類別(僅以如上所述的圖像類別採用五種類別為例，也可為更高或更低種)的分類結果作為多種信任分數，而非僅有單一的分類結果。In contrast, the image object classification method of the above-mentioned embodiment of the present invention adopts a plurality of binary classifiers, and each classifier only classifies two categories (for example, when it needs to be divided into 5 categories, 10 categories can be used. device), its advantage is at least that: the image object classification method of the above-mentioned embodiment of the present invention can know which of all categories are easy to distinguish and which are not easy to distinguish through each binary classification result, and add a full connection at the back The layer performs feature integration training, for example, the classification results of several categories (only five categories are used as an example for the image category as described above, or higher or lower) can be presented as a variety of trust scores in the output. , rather than a single classification result.
對比地，如果要採用二元分類器進行該非多重分類的多分類任務，其做法雖可在二元分類器的結果再額外加上一個決策機制，但此決策機制通常需要額外設定門檻值來決定最終結果，導致非客觀的人為介入成分較多。In contrast, if a binary classifier is to be used for the non-multi-classification multi-classification task, although an additional decision-making mechanism can be added to the results of the binary classifier, this decision-making mechanism usually requires an additional threshold value to decide. As a result, there are many non-objective human intervention components.
反觀地，本發明上述實施例的圖像物件分類方法則可使用全連接層代替決策機制，所有的決策門檻值都可透過學習而客觀地獲得，可以有效避免人為介入進行調整門檻值的影響。In contrast, the image object classification method of the above-mentioned embodiment of the present invention can use a fully connected layer to replace the decision-making mechanism, and all decision thresholds can be obtained objectively through learning, which can effectively avoid the influence of human intervention to adjust the thresholds.
表1 輸入資料表
舉例而言，將本發明上述實施例採用多重分類器的圖像物件分類方法，與另一種採用非多重分類器的方法進行實驗比較，例如：該非多重分類器採用一個輸入端、一個卷積神經網路(CNN)模型及一個輸出端；本發明上述實施例採用一個輸入端、十個二元分類器、一個全連接網路及一個輸出端，用於測試的輸入資料(諸如特性曲線、電路圖、腳位圖、時脈圖及三視圖)，可如上表1所示；而測試結果(諸如模型名稱、子模型、閾值、精度及訓練終止條件)，可如下表2所示。For example, compare the image object classification method using multiple classifiers in the above embodiment of the present invention with another method using non-multi-classifiers. For example, the non-multi-classifier uses one input, one convolution Network (CNN) model and an output terminal; the above-mentioned embodiment of the present invention adopts an input terminal, ten binary classifiers, a fully connected network and an output terminal, and the input data (such as characteristic curves, circuit diagrams) for testing , pin map, clock map, and three views), as shown in Table 1 above; and the test results (such as model name, sub-model, threshold, accuracy, and training termination conditions), as shown in Table 2 below.
表2 測試結果列表
從表2可知，本發明的圖像物件分類方法、系統、電腦程式產品及電腦可讀取紀錄媒體採用上述多重分類的分類效能，明顯優於非多重分類的分類效能。It can be seen from Table 2 that the image object classification method, system, computer program product and computer-readable recording medium of the present invention adopt the above-mentioned multi-classification classification performance, which is significantly better than that of non-multi-classification.
以下舉例說明本發明上述實施例的應用實例，作為瞭解本發明優點的應用示例，惟不以此為限。The following examples illustrate the application examples of the above-mentioned embodiments of the present invention, which are used as application examples to understand the advantages of the present invention, but are not limited thereto.
舉例而言，如第5A圖所示，其為一圖檔中含有諸多特性曲線區域C1、C2、C3、C4、C5及C6用於圖像物件分類的一示例，在此示例中，可有效將諸多特性曲線分類；如第5B圖所示，其為一圖檔中含有一電路圖區域K、一表格區域B及一文字區域T1用於圖像物件分類的一示例，在此示例中，可有效對電路圖進行分類；如第5C圖所示，其為一圖檔中含有諸多電子零件的腳位圖區域V1、V2、V3、V4及一文字區域T2用於圖像物件分類的一示例，在此示例中，可有效對腳位圖進行分類；如第5D圖所示，其為一圖檔中含有一訊號時序圖區域P用於圖像物件分類的一示例，在此示例中，可有效對訊號時序圖進行分類；如第5E圖所示，其為一圖檔中含有諸多電子零件的方向視圖區域Va、Vb、Vc、Vd、Ve、一表格區域Ba及一文字區域Ta用於圖像物件分類的一示例，在此示例中，可有效對諸多電子零件的方向視圖進行分類。從而，本發明的圖像物件分類方法、系統、電腦程式產品及電腦可讀取紀錄媒體，確實可以對數位文件中分布的單種圖像物件進行分類。For example, as shown in Fig. 5A, it is an example of a file containing many characteristic curve regions C1, C2, C3, C4, C5 and C6 for image object classification. In this example, it can be effectively Classify many characteristic curves; as shown in Figure 5B, it is an example of a drawing file containing a circuit diagram area K, a table area B and a text area T1 for image object classification. In this example, it can be effectively Classify the circuit diagram; as shown in Figure 5C, it is an example of the pin map areas V1, V2, V3, V4 and a text area T2 that contain many electronic parts in a drawing file for classifying image objects, here In the example, the pin map can be effectively classified; as shown in FIG. 5D, which is an example of a map file containing a signal timing diagram area P for image object classification, in this example, it can effectively classify the The signal timing diagram is classified; as shown in Figure 5E, it is a direction view area Va, Vb, Vc, Vd, Ve, a table area Ba, and a text area Ta containing many electronic components in a picture file for image objects. An example of classification, where directional views of many electronic parts can be effectively classified. Therefore, the image object classification method, system, computer program product and computer-readable recording medium of the present invention can indeed classify a single image object distributed in a digital file.
此外，如第6A圖所示，其為一圖檔中含有諸多電子零件的方向視圖區域Vf、Vg、Vh與特性曲線區域Ca混搭諸多表格區域Bb、Bc、Bd及諸多文字區域Tb、Tc用於圖像物件分類的一示例，在此示例中，可有效對方向視圖與特性曲線圖之多重類別進行分類，例如其順序可被配置為：先找出方向視圖，再找出特性曲線圖，惟不以此為限；如第6B圖所示，其為一圖檔中含有電子零件的腳位圖區域Vi與特性曲線區域Cb、Cc混搭諸多表格區域Be、Bf及文字區域Td用於圖像物件分類的一示例，在此示例中，腳位圖與特性曲線圖的類別都可被找出，例如其順序可被配置為：先找出腳位圖，再找出特性曲線圖，惟不以此為限。從而，本發明的圖像物件分類方法、系統、電腦程式產品及電腦可讀取紀錄媒體，確實可以對數位文件中分布的多種圖像物件進行分類。In addition, as shown in FIG. 6A , it is used to mix and match the directional view areas Vf, Vg, Vh and the characteristic curve area Ca, which contain many electronic parts, with many table areas Bb, Bc, Bd and many text areas Tb, Tc. In an example of image object classification, in this example, multiple categories of directional views and characteristic curves can be effectively classified. For example, the order can be configured as: first find the directional views, and then find the characteristic curve, However, it is not limited to this; as shown in Figure 6B, it is a map file containing the pin map area Vi of electronic parts and the characteristic curve areas Cb and Cc. Like an example of object classification, in this example, both the pin map and the characteristic curve category can be found, for example, the order can be configured as: first find the pin map, then find the characteristic curve, only Not limited to this. Therefore, the image object classification method, system, computer program product and computer-readable recording medium of the present invention can indeed classify a variety of image objects distributed in digital files.
因此，本發明的圖像物件分類方法、系統、電腦程式產品及電腦可讀取紀錄媒體，通過將輸入的圖像物件進行數種二元分類化特徵抽取處理、特徵抽象化處理及特徵整合處理，可以針對物件與類別之間的關係為一對一、多對一、一對多及多對多等多種分類情境進行處理，可以有效輸出多種分類結果作為信任分數，而非僅有單一的分類結果。Therefore, the image object classification method, system, computer program product and computer-readable recording medium of the present invention perform several binary classification feature extraction processes, feature abstraction processes, and feature integration processes on the input image objects. , which can deal with various classification situations such as one-to-one, many-to-one, one-to-many, and many-to-many relationships between objects and categories, and can effectively output multiple classification results as trust scores instead of only a single classification. result.
綜上所述，本發明的圖像物件分類方法、系統、電腦程式產品及電腦可讀取紀錄媒體，通過提供一圖檔，該圖檔包括至少一圖像物件；對該圖像物件進行數種二元分類化特徵抽取處理，以取得在類別上互相獨立的數個第一結果；將該數個第一結果以基於串接的降維方式進行結合，並對被結合之該數個第一結果以全連接方式進行特徵抽象化處理，以取得一第二結果；及對該數個第一結果及該第二結果進行特徵整合處理，以取得一分類結果。藉由上述物件分類過程，能夠輸出圖像物件歸屬於多種類別的隱含資訊，有利適用於含有多種圖像類別的數位文件。To sum up, in the image object classification method, system, computer program product and computer-readable recording medium of the present invention, by providing an image file, the image file includes at least one image object; A binary classification feature extraction process to obtain several first results that are independent of each other in category; combine the several first results in a dimensionality reduction method based on concatenation, and analyze the combined number of the first results. A result is subjected to feature abstraction processing in a fully connected manner to obtain a second result; and feature integration processing is performed on the plurality of first results and the second results to obtain a classification result. Through the above-mentioned object classification process, the implicit information that the image objects belong to various categories can be output, which is advantageously applicable to digital files containing various image categories.
雖然本發明已以較佳實施例揭露，然其並非用以限制本發明，任何熟習此項技藝之人士，在不脫離本發明之精神和範圍內，當可作各種更動與修飾，因此本發明之保護範圍當視後附之申請專利範圍所界定者為準。Although the present invention has been disclosed with preferred embodiments, it is not intended to limit the present invention. Any person skilled in the art can make various changes and modifications without departing from the spirit and scope of the present invention. Therefore, the present invention The scope of protection shall be determined by the scope of the appended patent application.
1:處理裝置 11:分類模組 12:格式轉換模組 A1:零件分類單元 A2:零件視圖辨識單元 B:表格區域 Ba:表格區域 Bb:表格區域 Bc:表格區域 Bd:表格區域 Be:表格區域 Bf:表格區域 C1~C6:特性曲線區域 Ca:特性曲線區域 Cb:特性曲線區域 Cc:特性曲線區域 D:外部資料 K:電路圖區域 P:訊號時序圖區域 R:分類結果 S1~S6:步驟 T1:文字區域 T2:文字區域 Ta:文字區域 Tb:文字區域 Tc:文字區域 Td:文字區域 V1~V4:電子零件的腳位圖區域 Va:電子零件的方向視圖區域 Vb:電子零件的方向視圖區域 Vc:電子零件的方向視圖區域 Vd:電子零件的方向視圖區域 Ve:電子零件的方向視圖區域 Vf:電子零件的方向視圖區域 Vg:電子零件的方向視圖區域 Vh:電子零件的方向視圖區域 Vi:電子零件的腳位圖區域 X:多重分類器 Y:非多重分類器 1: Processing device 11: Classification module 12: Format conversion module A1: Parts classification unit A2: Part View Recognition Unit B: table area Ba: table area Bb: table area Bc: table area Bd: table area Be: table area Bf: table area C1~C6: characteristic curve area Ca: characteristic curve area Cb: characteristic curve area Cc: characteristic curve area D: External information K: circuit diagram area P: Signal timing diagram area R: Classification result S1~S6: Steps T1: Text area T2: Text area Ta: text area Tb: text area Tc: text area Td: text area V1~V4: pin map area of electronic parts Va: Orientation view area for electronic parts Vb: Orientation view area of electronic parts Vc: Orientation view area for electronic parts Vd: Orientation view area for electronic parts Ve: Orientation view area for electronic parts Vf: Orientation view area for electronic parts Vg: Orientation view area for electronic parts Vh: Orientation view area for electronic parts Vi: Footprint area of electronic parts X: Multiple Classifiers Y: non-multi-classifier
［第1圖］：本發明實施例之圖像物件分類系統的方塊示意圖。 ［第2圖］：本發明實施例之圖像物件分類方法的流程示意圖。 ［第3圖］：本發明實施例採用之一種多重分類器的示意圖。 ［第4圖］：本發明實施例未採用之另一種非多重分類器的示意圖。 ［第5A圖］：本發明實施例用於單種圖像物件分類的範例圖像示意圖(一)。 ［第5B圖］：本發明實施例用於單種圖像物件分類的範例圖像示意圖(二)。 ［第5C圖］：本發明實施例用於單種圖像物件分類的範例圖像示意圖(三)。 ［第5D圖］：本發明實施例用於單種圖像物件分類的範例圖像示意圖(四)。 ［第5E圖］：本發明實施例用於單種圖像物件分類的範例圖像示意圖(五)。 ［第6A圖］：本發明實施例用於多種圖像物件分類的範例圖像示意圖(一)。 ［第6B圖］：本發明實施例用於多種圖像物件分類的範例圖像示意圖(二)。 [FIG. 1]: A block diagram of an image object classification system according to an embodiment of the present invention. [Fig. 2]: A schematic flowchart of a method for classifying image objects according to an embodiment of the present invention. [Fig. 3]: A schematic diagram of a multiple classifier used in an embodiment of the present invention. [FIG. 4]: A schematic diagram of another non-multiple classifier not used in the embodiment of the present invention. [Fig. 5A]: A schematic diagram (1) of an example image used for classifying a single image object according to an embodiment of the present invention. [Fig. 5B]: A schematic diagram (2) of an example image used for classifying a single image object according to an embodiment of the present invention. [Fig. 5C]: A schematic diagram (3) of an example image used for classifying a single image object according to an embodiment of the present invention. [FIG. 5D]: A schematic diagram (4) of an example image used for classifying a single image object according to an embodiment of the present invention. [Fig. 5E]: A schematic diagram (5) of an example image used for classifying a single image object according to an embodiment of the present invention. [Fig. 6A]: A schematic diagram (1) of an example image used for classifying various image objects according to an embodiment of the present invention. [Fig. 6B]: A schematic diagram (2) of an example image used for classifying various image objects according to an embodiment of the present invention.
S1~S6:步驟 S1~S6: Steps
Claims (10)
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| TW110100510A TW202228066A (en) | 2021-01-06 | 2021-01-06 | Image object classification method, system, computer program product, and computer readable medium |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| TW110100510A TW202228066A (en) | 2021-01-06 | 2021-01-06 | Image object classification method, system, computer program product, and computer readable medium |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| TW202228066A true TW202228066A (en) | 2022-07-16 |
Family
ID=83436899
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| TW110100510A TW202228066A (en) | 2021-01-06 | 2021-01-06 | Image object classification method, system, computer program product, and computer readable medium |
Country Status (1)
| Country | Link |
|---|---|
| TW (1) | TW202228066A (en) |
-
2021
- 2021-01-06 TW TW110100510A patent/TW202228066A/en unknown
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| WO2021073417A1 (en) | Expression generation method and apparatus, device and storage medium | |
| AU2016259337B2 (en) | Modeling semantic concepts in an embedding space as distributions | |
| US11328172B2 (en) | Method for fine-grained sketch-based scene image retrieval | |
| US9639806B2 (en) | System and method for predicting iconicity of an image | |
| CN110674741B (en) | Gesture recognition method in machine vision based on double-channel feature fusion | |
| CN112148916A (en) | Cross-modal retrieval method, device, equipment and medium based on supervision | |
| CN112801146B (en) | Target detection method and system | |
| CN109993197A (en) | A kind of zero sample multi-tag classification method based on the end-to-end example differentiation of depth | |
| CN107808129A (en) | A kind of facial multi-characteristic points localization method based on single convolutional neural networks | |
| Sun et al. | Traffic Sign Recognition Method Integrating Multi-Layer Features and Kernel Extreme Learning Machine Classifier. | |
| WO2022247112A1 (en) | Task processing method and apparatus, device, storage medium, computer program, and program product | |
| CN109508625A (en) | Emotional data analysis method and device | |
| CN111460782A (en) | Information processing method, device and equipment | |
| KR20130059035A (en) | Apparatus and method for completing automatically scenario continuity using sketch, and system for making animation using thereof | |
| CN112037239A (en) | Text guidance image segmentation method based on multi-level explicit relation selection | |
| US20230060459A1 (en) | Image object classification optimizing method, system and computer readable medium | |
| US11947634B2 (en) | Image object classification method, system and computer readable medium | |
| Wei et al. | Color image retrieval based on quaternion and deep features | |
| Sunitha et al. | Novel content based medical image retrieval based on BoVW classification method | |
| Jin et al. | Image classification based on pLSA fusing spatial relationships between topics | |
| TW202228066A (en) | Image object classification method, system, computer program product, and computer readable medium | |
| CN115205877A (en) | Irregular typesetting invoice document layout prediction method and device and storage medium | |
| CN113987170A (en) | Multi-label text classification method based on convolutional neural network | |
| CN113837263A (en) | Gesture image classification method based on feature fusion attention module and feature selection | |
| GB2546368A (en) | Modelling semantic concepts in an embedding space as distributions |