TW202016757A

TW202016757A - Commodity image search system and method thereof

Info

Publication number: TW202016757A
Application number: TW107137604A
Authority: TW
Inventors: 杜師偉
Original assignee: 光點開發股份有限公司
Priority date: 2018-10-24
Filing date: 2018-10-24
Publication date: 2020-05-01

Abstract

A commodity image search system includes a commodity data base, a consumer terminal and a processor. A plurality of commodity data is saved in the commodity data base. The commodity data base includes a commodity image and a image analysis information. The consumer terminal includes an input module. The input module is configured to input a subject image. The processor is configured to receive the subject image. After the processor receives the subject image, the subject image would be transmitted to external image identification module and get an analysis information of the subject image. The processor compares the analysis information of the subject image with the image analysis information and selects the commodity data has the image analysis information like as the analysis information of the subject image, and the selected commodity data would be transmitted to the consumer terminal.

Description

商品圖像比較搜尋系統與其方法Commodity image comparison search system and method

一種商品搜尋系統，特別是一種商品圖像比較搜尋系統。A commodity search system, especially a commodity image comparison search system.

圖像辨識是現在資訊技術的一種，透過輸入圖象讓電腦系統辨識該圖像為何。而且，圖像辨識的技術也廣泛應用在搜尋引擎上，讓網路的使用者可透過輸入圖辨來搜尋所要的資訊。然而，目前圖像辨識的精確程度仍有限，多半僅能提供大方向的搜尋結果。倘若使用者想要搜尋更細節的項目，例如想搜尋款式相似的服飾或產品，由於多數圖形搜尋引擎無法切確辨識產品款式，就很難搜尋到有用資訊。因此，如何讓使用者能夠更容易地以產品搜尋類似產品，便是本領域具通常知識者值得去思量的。Image recognition is one of the current information technologies. It allows the computer system to recognize the image by inputting the image. Moreover, the technology of image recognition is also widely used in search engines, allowing users on the Internet to search for the desired information by entering image recognition. However, the accuracy of image recognition is still limited, and most of them can only provide general search results. If users want to search for more detailed items, such as clothing or products with similar styles, because most graphic search engines cannot accurately identify product styles, it is difficult to search for useful information. Therefore, how to make it easier for users to search for similar products by products is worth considering for those with ordinary knowledge in the field.

本發明提供一種商品圖像比較搜尋系統，可由消費者輸入圖像至系統中，由系統為消費者挑出最接近該圖像的商品資料。本發明提供一種商品圖像比較搜尋系統，包括一商品資料庫、一消費端及一處理器。商品資料庫儲存有多筆商品資料，該商品資料還包括一商品圖像與一商品圖像分析資訊。消費端包括一輸入模組，適於輸入一標的物圖像。處理器適於接收該標的物圖像。其中，該處理器接收該標的物圖像後，將該標的物圖像傳送至外部的一圖像識別模組，並取得一標的物圖像分析資訊，該處理器比對該標的物圖像分析資訊與該商品圖像分析資訊，並選出該商品圖像分析資訊與該標的物圖像分析資訊相似的該商品資料，將該商品資料傳送至該消費端。上述之商品圖像比較搜尋系統，其中，該處理器將該商品圖像傳送至該圖像識別模組，取得該商品圖像分析資訊。上述之商品圖像比較搜尋系統，其中，還包括一商務端，適於輸入該商品資訊。上述之商品圖像比較搜尋系統，其中，該處理器自動將該商務端輸入的該商品資料傳送至該圖像識別模組。上述之商品圖像比較搜尋系統，其中，該商品圖像分析資訊與該標的物圖像分析資訊為JSON格式的資料。本發明還提供一種商品圖像比對的方法，包括： S10：將一商品圖像傳送至外部的一圖像識別模組； S20：取得多個標籤資訊、多個網頁資訊、多個文字資訊與多個色彩資訊； S30：將該標籤資訊與該網頁資訊進行交叉運算，取得多個精準分數、多個模糊分數，從該文字資訊取得多個文字分數，從該色彩資訊取得多個色彩分數； S40：將該商品圖像、該精確分數、該模糊分數、該文字分數與該色彩分數保存至一商品資料庫； S50：若該商品圖像是由一消費端輸入，將步驟S30中所取得的將該精確分數、該模糊分數、該文字分數、該色彩分數與該商品資料庫中的將該精確分數、該模糊分數、該文字分數、該色彩分數進行交叉演算比對；及 S60：從商品資料庫中選出媒合的該商品圖像並傳送至該消費端。上述之商品圖像比對的方法，其中，在步驟S30中，是從該標籤資訊與該網頁資訊中挑出具有相同元素的該標籤資訊與該網頁資訊計算該精確分數。上述之商品圖像比對的方法，其中，在步驟S30中，是對該色彩資訊進行16位元減法取絕對值計算該色彩分數。The invention provides a commodity image comparison and search system, in which an image can be input by a consumer into the system, and the system can pick out the commodity material closest to the image for the consumer. The invention provides a commodity image comparison search system, which includes a commodity database, a consumer and a processor. The commodity database stores multiple pieces of commodity data, and the commodity data also includes a commodity image and a commodity image analysis information. The consumer terminal includes an input module suitable for inputting an object image. The processor is adapted to receive the target object image. After receiving the target object image, the processor transmits the target object image to an external image recognition module, and obtains a target object image analysis information. The processor compares the target object image Analyze the information and the product image analysis information, and select the product data that is similar to the target object image analysis information, and send the product data to the consumer. In the aforementioned commodity image comparison search system, wherein the processor transmits the commodity image to the image recognition module to obtain the commodity image analysis information. The above product image comparison search system, which also includes a business end, is suitable for inputting the product information. In the above commodity image comparison and search system, the processor automatically transmits the commodity data input from the business end to the image recognition module. In the above commodity image comparison search system, the commodity image analysis information and the target object image analysis information are data in JSON format. The invention also provides a method for comparing commodity images, including: S10: transmitting a commodity image to an external image recognition module; S20: acquiring multiple tag information, multiple web page information, and multiple text information And multiple color information; S30: cross-computing the label information and the web page information to obtain multiple precision scores and multiple fuzzy scores, obtain multiple text scores from the text information, and obtain multiple color scores from the color information ; S40: save the commodity image, the precise score, the blur score, the text score and the color score to a commodity database; S50: if the commodity image is input by a consumer, the step S30 The obtained accurate score, the fuzzy score, the text score, and the color score are cross-computed and compared with the accurate score, the fuzzy score, the text score, and the color score in the product database; and S60: The matching image of the product is selected from the product database and sent to the consumer. In the above method for comparing commodity images, in step S30, the tag information and the web page information having the same elements are selected from the tag information and the web page information to calculate the accurate score. In the above method for comparing commodity images, in step S30, 16-bit subtraction is performed on the color information to obtain the absolute value to calculate the color score.

本發明提供一種商品圖像比較搜尋系統，使用者可輸入商品的圖像，由商品圖像比較搜尋系統找出最相似的產品供使用者參考選購。請參閱圖1，圖1所繪示為本發明的商品圖像比較搜尋系統之架構圖。本發明之商品圖像比較搜尋系統100包括一商品資料庫130、一消費端120、一處理器110與一商務端140。商品資料庫130中儲存有多筆商品資料，商品資料還包括一商品圖像與商品圖像分析資訊。其中商品圖像為商品本身的影像，例如商品照片。而商品圖像分析資訊則是經由將商品圖像傳送至外部之一圖像識別模組10所取得的資訊，其取得方式容後再述。消費端120則是消費者使用的終端裝置，例如智慧型手機或個人電腦，消費端120適於輸入一標的物圖像。標的物圖像為消費者所欲搜尋的商品圖像，例如為商品的照片，如服飾配件等照片。消費端120所輸入的標的物圖像則會被傳送至處理器110。處理器110為商品圖像比較搜尋系統100的運算中心，適於運算與比較各項資訊。在本實施例中，處理器110在接收標的物圖像之後，會將標的物圖像傳送至圖像識別模組10。圖像識別模組10是一個外部的識別系統，例如為Google的雲端人工智慧與機械學習系統（AI & Machine Learning Products）。圖像識別模組10在被輸入標的物圖像檔案之後，會針對標的物圖像產生標的物圖像的識別數據。而處理器110會接收這些標的物圖像的識別數據，並加以運算取得標的物分析資訊。接下來，處理器110會進一步比較標的物分析資訊與商品圖像分析資訊。並且從商品資料庫130中挑選與標的物分析資訊相符的商品資料，並將商品資料傳送至消費端120。至此，也就是消費者經由消費端120上傳產品的照片，經由商品圖像比較搜尋系統100提供相似的產品，並傳送至消費端120供消費者參考。商務端140則是商品提供者所使用的終端裝置，商品提供者可經由商務端140將商品資料輸入至商品圖像比較搜尋系統100。經由商務端140輸入的商品資料與商品圖像由處理器110接收，處理器110將商品圖像傳送至圖像識別模組10，並從圖像識別模組10接收商品圖像的分析數據，並且經由商品資料的分析數據計算出商品圖像分析資訊，而商品圖像與商品圖像分析資訊會被儲存至商品資料庫130。也就是說，商務端140是供商品提供者上傳商品資料至商品資料庫130，進一步建構商品圖像比較搜尋系統100中可供消費者參考的資料。在一實施例中，商品圖像比較搜尋系統100會自動將商務端140所提供的商品資料傳送至圖像識別模組10，並且加以分析並補充商品資料庫130中的商品資料。例如，商務端140上傳欲銷售之商品資料後（如於電子商務網上建立產品銷售頁），商品圖像比較搜尋系統100便會自動為其擴充商品資料庫130。請參閱圖2A，圖2A所繪示為圖像識別模組的示意圖。當圖像識別模組10接收到圖像之後，例如標的物圖像或商品圖像，會產生多個標籤資訊、多個網頁資訊、多個文字資訊與多個色彩資訊，並且給予這些資訊一個評分或百分比。而標籤資訊、網頁資訊、文字資訊與色彩資訊會由處理器110接收。處理器110會針對標籤資訊與網頁資訊中挑出具有相同元素的標籤資訊或網頁資訊，例如圖2A中的「西裝外套」與「無尾禮服」的元素。請參閱圖2B，圖2B所繪示為精確分數與模糊分數的示意圖。從標籤資訊與網頁資訊中挑出具有相同元素的標籤資訊或網頁資訊之後，處理器110會將具有相同元素的標籤資訊或網頁資訊的百分比或評分相加除於二，算出精確分數。而只有單一元素的標籤資訊或網頁資訊則將其百分比或評分直接除於二，算出模糊分數。此外，處理器110還會從文字資訊計算出文字分數；從色彩資訊中計算出色彩分數，並且是對色彩資訊進行16位元減法取絕對值計算色彩分數。在本實施例中，處理器110是將標籤資訊、網頁資訊、文字資訊與色彩資訊轉換成JSON格式的程式碼進行交叉運算。所計算出的精確分數、模糊分數、文字分數與色彩分數，都會用跟消費端130所輸入的標的物圖像進行比對。請參閱圖3，圖3所繪示為商品圖像比對的方法。首先，將產品圖像輸入至圖像辨識模組10（步驟S10），並且取得產品圖像的分析資訊（步驟S20）。步驟S10與步驟S20即是利用圖像辨識模組10取得商品圖像的標籤資訊、網頁資訊、文字資訊與色彩資訊，作為計算分數的根據。接下來，經過交叉運算取得商品圖像的各項分數（步驟S30），包括精確分數、模糊分數、文字分數與色彩分數。精確分數、模糊分數、文字分數與色彩分數的計算方式如上述圖2A與圖2B之說明，由標籤資訊與網頁資訊交叉運算出精確分數與模糊分數，從文字資訊獲得文字分數，從色彩資訊獲得色彩分數。接下來，將商品圖像與其對應的精確分數、模糊分數、文字分數與色彩分數傳送至商品資料庫保存130（步驟S40）。至此，步驟S10至步驟S40例如由商務端140提供產品圖像與產品資料，並且經由分析產品圖進一步建立產品資料庫130中產品資料與對應的精確分數、模糊分數、文字分數與色彩分數。然後，若產品圖像是由消費端120所輸入，在步驟S30中所取得的精確分數、模糊分數、文字分數、色彩分數與商品資料庫130中的精確分數、模糊分數、文字分數、色彩分數進行交叉演算比較（步驟S50）。接下來，從商品資料庫中選出媒合的該商品圖像並傳送至消費端120（步驟S60）。也就是說，將步驟S50中比較出商品資料庫130與消費端120所輸入的商品圖像最接近的商品資料傳送至消費端120。步驟S50至步驟S60則是消費者傳送商品圖像後，從商品資料庫130中選出最符合消費者需求的項目，並提供給消費者參考。本發明之商品圖像比較搜尋系統與其方法，可以讓消費者輸入隨機的商品圖像，利用外部的圖像辨識模組10產生分析資訊，並進一步由商品圖像比較搜尋系統100計算評分分數。在從商品資料庫130中挑出評分分數最接近的商品圖像供消費者參考。例如消費者輸入品牌不詳的服裝圖像，商品圖像比較搜尋系統100便可從商品資料庫130中挑出最接近消費者所輸入的服裝圖像之商品資料，可大幅提高消費者尋找商品的便利性，尤其在消費者無法得知其需求商品的具體名稱或敘述時，只要提供商品的圖像即可。另外，值得注意的是，上述雖以服裝作為商品的實施例，但本發明具有通常知識者應可知商品圖像比較搜尋系統100並不限於用在辨識服裝，也可用於辨識其他商品。特別是本發明在較佳實施例中，圖像辨識模組10是採用Google的雲端人工智慧與機械學習系統（Google Cloud AI & Machine Learning Products），該系統經過Google大量的數據不斷的學習，因此能夠提供準確度極高的圖像識別能力，並且將識別結果轉化為常用的程式碼，讓開發者得以運用。而本發明的商品圖像比較搜尋系統100便是讓成衣業者或其他商品的經銷商能夠利用Google的雲端人工智慧與機械學習系統強大的圖像識別能力，提供更好的商品搜尋服務。本發明說明如上，然其並非用以限定本創作所主張之專利權利範圍。其專利保護範圍當視後附之申請專利範圍及其等同領域而定。凡本領域具有通常知識者，在不脫離本專利精神或範圍內，所作之更動或潤飾，均屬於本創作所揭示精神下所完成之等效改變或設計，且應包含在下述之申請專利範圍內。The invention provides a commodity image comparison and search system. The user can input the image of the commodity, and the commodity image comparison and search system finds the most similar products for the user to refer to and purchase. Please refer to FIG. 1, which is a structural diagram of a product image comparison search system of the present invention. The product image comparison search system 100 of the present invention includes a product database 130, a consumer 120, a processor 110, and a business 140. The commodity database 130 stores a plurality of commodity data, and the commodity data further includes a commodity image and commodity image analysis information. The product image is an image of the product itself, such as a product photo. The product image analysis information is obtained by transmitting the product image to an external image recognition module 10, and the method of obtaining the content will be described later. The consumer terminal 120 is a terminal device used by consumers, such as a smart phone or a personal computer. The consumer terminal 120 is suitable for inputting a target object image. The target object image is the image of the product that the consumer wants to search, for example, a photo of the product, such as a photo of clothing accessories. The target image input by the consumer 120 is sent to the processor 110. The processor 110 is a computing center of the commodity image comparison search system 100, and is suitable for computing and comparing various pieces of information. In this embodiment, after receiving the target object image, the processor 110 transmits the target object image to the image recognition module 10. The image recognition module 10 is an external recognition system, such as Google's cloud artificial intelligence and machine learning products (AI & Machine Learning Products). After being input into the target object image file, the image recognition module 10 generates identification data of the target object image for the target object image. The processor 110 receives the identification data of these target object images and performs calculation to obtain the target object analysis information. Next, the processor 110 will further compare the subject matter analysis information with the commodity image analysis information. And select the commodity data that matches the analysis information of the object from the commodity database 130, and send the commodity data to the consumer 120. So far, consumers upload photos of products through the consumer 120, provide similar products through the product image comparison and search system 100, and send them to the consumer 120 for reference by consumers. The commercial terminal 140 is a terminal device used by the commodity provider. The commodity provider can input the commodity information to the commodity image comparison and search system 100 via the commercial terminal 140. The commodity data and commodity image input via the business terminal 140 are received by the processor 110, and the processor 110 transmits the commodity image to the image recognition module 10, and receives the analysis data of the commodity image from the image recognition module 10. Furthermore, the product image analysis information is calculated from the analysis data of the product data, and the product image and the product image analysis information are stored in the product database 130. That is to say, the business end 140 is for the commodity provider to upload commodity data to the commodity database 130, and further constructs the commodity image comparison search system 100 for reference by consumers. In one embodiment, the commodity image comparison and search system 100 automatically sends the commodity data provided by the commercial terminal 140 to the image recognition module 10, and analyzes and supplements the commodity data in the commodity database 130. For example, after the commercial terminal 140 uploads the product information to be sold (such as creating a product sales page on the e-commerce website), the product image comparison search system 100 will automatically expand the product database 130 for it. Please refer to FIG. 2A, which is a schematic diagram of an image recognition module. After the image recognition module 10 receives the image, such as the target object image or the product image, it will generate multiple tag information, multiple web page information, multiple text information, and multiple color information, and give these information a Score or percentage. The tag information, web page information, text information, and color information are received by the processor 110. The processor 110 will pick out tag information or web page information having the same elements from the tag information and the web page information, such as the elements of "blazer" and "tuxedo" in FIG. 2A. Please refer to FIG. 2B, which is a schematic diagram of an accurate score and a fuzzy score. After picking out tag information or webpage information with the same elements from the tag information and webpage information, the processor 110 adds the percentage or score of the tag information or webpage information with the same elements to two to calculate an accurate score. For label information or webpage information with only a single element, the percentage or score is directly divided by two to calculate the fuzzy score. In addition, the processor 110 also calculates the text score from the text information; calculates the color score from the color information, and performs a 16-bit subtraction on the color information to take the absolute value to calculate the color score. In this embodiment, the processor 110 converts the label information, web page information, text information, and color information into JSON format code for cross calculation. The calculated exact score, blur score, text score and color score will be compared with the target object image input by the consumer 130. Please refer to FIG. 3, which illustrates a method for comparing commodity images. First, the product image is input to the image recognition module 10 (step S10), and the analysis information of the product image is obtained (step S20). Steps S10 and S20 use the image recognition module 10 to obtain the label information, web page information, text information, and color information of the product image as the basis for calculating the score. Next, each score of the product image is obtained through cross calculation (step S30), including the precise score, the blur score, the text score, and the color score. The calculation method of exact score, fuzzy score, text score and color score is as described above in FIG. 2A and FIG. 2B. The precise score and blur score are calculated from the label information and the web page information. The text score is obtained from the text information and the color information is obtained from the color information. Color score. Next, the product image and its corresponding accurate score, blur score, text score and color score are sent to the product database storage 130 (step S40). So far, from step S10 to step S40, for example, the business end 140 provides product images and product data, and further creates product data in the product database 130 and corresponding precision scores, blur scores, text scores, and color scores by analyzing the product map. Then, if the product image is input by the consumer 120, the precise score, blur score, text score, color score obtained in step S30 and the precise score, blur score, text score, color score in the product database 130 A cross calculation comparison is performed (step S50). Next, the matched product image is selected from the product database and transmitted to the consumer 120 (step S60). In other words, the product data that compares the product images input by the product database 130 and the consumer 120 in step S50 is closest to the consumer 120. Steps S50 to S60 are that after the consumer transmits the commodity image, the item that best meets the consumer's needs is selected from the commodity database 130 and provided to the consumer for reference. The product image comparison search system and method of the present invention allow consumers to input random product images, use external image recognition modules 10 to generate analysis information, and further calculate the scores by the product image comparison search system 100. The image of the product with the closest rating score is selected from the product database 130 for the consumer's reference. For example, if a consumer enters a clothing image with an unknown brand, the product image comparison search system 100 can select the product information closest to the clothing image entered by the consumer from the product database 130, which can greatly increase the consumer's search for products Convenience, especially when consumers cannot know the specific name or description of the product they need, just provide an image of the product. In addition, it is worth noting that although the above-mentioned embodiment uses clothing as a product, those with ordinary knowledge in the present invention should understand that the product image comparison search system 100 is not limited to identifying clothing, but can also be used to identify other products. In particular, in a preferred embodiment of the present invention, the image recognition module 10 uses Google’s cloud artificial intelligence and machine learning products (Google Cloud AI & Machine Learning Products). The system continuously learns through Google’s massive data, so It can provide highly accurate image recognition capabilities, and convert the recognition results into commonly used code for developers to use. The product image comparison search system 100 of the present invention is to enable garment manufacturers or distributors of other products to use Google's cloud artificial intelligence and the powerful image recognition capabilities of the machine learning system to provide better product search services. The description of the present invention is as above, but it is not intended to limit the scope of patent rights claimed in this creation. The scope of patent protection depends on the scope of the attached patent application and its equivalent fields. Anyone with ordinary knowledge in this field who makes changes or retouching without departing from the spirit or scope of this patent belongs to the equivalent change or design completed under the spirit disclosed in this creation, and shall be included in the scope of the following patent application Inside.

10:圖像識別模組100:商品圖像比較搜尋系統110:處理器120:消費端130:商品資料庫140:商務端S10~S60:流程圖步驟10: Image recognition module 100: Commodity image comparison search system 110: Processor 120: Consumer 130: Commodity database 140: Business S10~S60: Flow chart steps

圖1所繪示為本發明的商品圖像比較搜尋系統之架構圖。圖2A所繪示為圖像識別模組的示意圖。圖2B所繪示為精確分數與模糊分數的示意圖。圖3所繪示為商品圖像比對的方法。FIG. 1 is a schematic diagram of a product image comparison search system of the present invention. FIG. 2A is a schematic diagram of an image recognition module. FIG. 2B is a schematic diagram of exact scores and fuzzy scores. Figure 3 illustrates a method for comparing commodity images.

10:圖像識別模組 10: Image recognition module

100:商品圖像比較搜尋系統 100: Product image comparison search system

110:處理器 110: processor

120:消費端 120: Consumer

130:商品資料庫 130: Commodity database

140:商務端 140: Business side

Claims

一種商品圖像比較搜尋系統，適於通訊連接至一圖像識別模組，包括：一商品資料庫，儲存有多筆商品資料，該商品資料還包括一商品圖像與一商品圖像分析資訊；一消費端，包括一輸入模組，適於輸入一標的物圖像；及一處理器，適於接收該標的物圖像；其中，該處理器接收該標的物圖像後，將該標的物圖像傳送至外部的一圖像識別模組，並取得一標的物圖像分析資訊，該處理器比對該標的物圖像分析資訊與該商品圖像分析資訊，並選出該商品圖像分析資訊與該標的物圖像分析資訊相似的該商品資料，將該商品資料傳送至該消費端。A commodity image comparison and search system, suitable for communication connection to an image recognition module, including: a commodity database storing multiple commodity data, the commodity data also includes a commodity image and a commodity image analysis information A consumer terminal, including an input module, suitable for inputting a target object image; and a processor, suitable for receiving the target object image; wherein, after the processor receives the target object image, the target object The object image is sent to an external image recognition module to obtain a target object image analysis information. The processor compares the target object image analysis information with the product image analysis information and selects the product image The commodity data whose analysis information is similar to the target object image analysis information is transmitted to the consumer.

如申請專利範圍第1項所述之商品圖像比較搜尋系統，其中，該處理器將該商品圖像傳送至該圖像識別模組，取得該商品圖像分析資訊。The product image comparison search system as described in item 1 of the patent application scope, wherein the processor transmits the product image to the image recognition module to obtain the product image analysis information.

如申請專利範圍第1項所述之商品圖像比較搜尋系統，其中，還包括一商務端，適於輸入該商品資訊。The product image comparison search system as described in item 1 of the patent application scope, which also includes a business end, is suitable for inputting the product information.

如申請專利範圍第3項所述之商品圖像比較搜尋系統，其中，該處理器自動將該商務端輸入的該商品資料傳送至該圖像識別模組。The commodity image comparison search system as described in item 3 of the patent application scope, wherein the processor automatically transmits the commodity data input from the business end to the image recognition module.

如申請專利範圍第1項所述之商品圖像比較搜尋系統，其中，該商品圖像分析資訊與該標的物圖像分析資訊為JSON格式的資料。The product image comparison search system as described in item 1 of the patent application scope, wherein the product image analysis information and the target object image analysis information are data in JSON format.

一種商品圖像比對的方法，包括： S10：將一商品圖像傳送至外部的一圖像識別模組； S20：取得多個標籤資訊、多個網頁資訊、多個文字資訊與多個色彩資訊； S30：將該標籤資訊與該網頁資訊進行交叉運算，取得多個精準分數、多個模糊分數，從該文字資訊取得多個文字分數，從該色彩資訊取得多個色彩分數； S40：將該商品圖像、該精確分數、該模糊分數、該文字分數與該色彩分數保存至一商品資料庫； S50：若該商品圖像是由一消費端輸入，將步驟S30中所取得的將該精確分數、該模糊分數、該文字分數、該色彩分數與該商品資料庫中的將該精確分數、該模糊分數、該文字分數、該色彩分數進行交叉演算比對；及 S60：從商品資料庫中選出媒合的該商品圖像並傳送至該消費端。A method for comparing commodity images, including: S10: sending a commodity image to an external image recognition module; S20: acquiring multiple tag information, multiple web page information, multiple text information, and multiple colors Information; S30: Cross-compute the tag information with the web page information to obtain multiple precision scores and multiple fuzzy scores, obtain multiple text scores from the text information, and obtain multiple color scores from the color information; S40: convert The commodity image, the precise score, the blur score, the text score and the color score are saved to a commodity database; S50: If the commodity image is input by a consumer, the The exact score, the fuzzy score, the text score, and the color score are cross-computed with the precise score, the fuzzy score, the text score, and the color score in the product database; and S60: from the product database The product image that is matched is selected and sent to the consumer.

如申請專利範圍第6項所述之商品圖像比對的方法，其中，在步驟S30中，是從該標籤資訊與該網頁資訊中挑出具有相同元素的該標籤資訊與該網頁資訊計算該精確分數。The method for comparing product images as described in item 6 of the patent application scope, wherein in step S30, the label information and the web page information with the same elements are selected from the label information and the web page information to calculate the Exact score.

如申請專利範圍第6項所述之商品圖像比對的方法，其中，在步驟S30中，是對該色彩資訊進行16位元減法取絕對值計算該色彩分數。The method for comparing commodity images as described in item 6 of the patent application scope, wherein in step S30, the color information is calculated by subtracting the absolute value of 16-bit subtraction of the color information.