TWI697844B - Visual artificial intelligence identification method and visual artificial intelligence identification system - Google Patents

Abstract

The invention relates to a visual artificial intelligence identification method and a visual artificial intelligence identification system. The visual artificial intelligence identification method includes: receiving image data of multiple objects to be identified; performing artificial intelligence identification on the image data of multiple objects to be identified to obtain an artificial intelligence identification result; judging the artificial intelligence identification result to obtain a Corresponding accuracy rate; based on the corresponding accuracy rate, decide whether to intervene a setting condition to the artificial intelligence recognition to perform artificial intelligence recognition again; and repeat the above artificial intelligence recognition, determine the artificial intelligence recognition result, and decide whether to intervene a setting condition To complete the identification of multiple objects to be identified.

Description

Visual artificial intelligence identification method and visual artificial intelligence identification system

The invention relates to a visual artificial intelligence identification method and a visual artificial intelligence identification system.

Xizhi uses artificial intelligence technology to identify image data, and inputs multiple image data for artificial intelligence identification. If the image data is too much and too complex, the artificial intelligence recognition may collapse and the recognition cannot be completed; or, if the image data that has completed the artificial intelligence training is re-input for artificial intelligence recognition, the result will be less stable , The result of artificial intelligence recognition is quite unstable, making the accuracy of artificial intelligence recognition quite low, and the result of conventional artificial intelligence recognition is quite unreliable.

In one embodiment, a visual artificial intelligence identification method includes: receiving image data of multiple objects to be identified; performing artificial intelligence identification on the image data of multiple objects to be identified to obtain artificial intelligence identification results; and judging artificial intelligence The recognition result to obtain a corresponding accuracy rate; based on the corresponding accuracy rate, decide whether to intervene a setting condition to the artificial intelligence recognition to perform the artificial intelligence recognition again; and repeat the above artificial intelligence recognition and judge the artificial intelligence recognition result and Decide whether to intervene in a step of setting conditions to complete the identification of multiple objects to be identified.

In one embodiment, a visual artificial intelligence recognition system includes: a receiving device, an artificial intelligence recognition device, a judging device, and an intervention device. The receiving device is used for receiving image data of a plurality of objects to be identified. The artificial intelligence recognition device is used for performing artificial intelligence recognition on the image data of a plurality of objects to be recognized to obtain artificial intelligence recognition results. The judging device is used to judge the artificial intelligence recognition result to obtain a corresponding accuracy rate. According to the corresponding accuracy rate, the intervention device decides whether to intervene a set condition to the artificial intelligence identification device to perform artificial intelligence identification again.

The visual artificial intelligence recognition method and the visual artificial intelligence recognition system of the present invention can be applied to recognize image data, such as credential data or medical image data, but are not limited to the above.

Fig. 1 shows a flowchart of the visual artificial intelligence identification method of the present invention. Figure 2 shows a block diagram of the visual artificial intelligence recognition system of the present invention. With reference to FIG. 1 and FIG. 2, please refer to step S11 to receive image data of multiple objects to be identified. In one embodiment, the visual artificial intelligence recognition system 20 of the present invention utilizes a receiving device 21 for receiving image data of multiple objects to be recognized. In one embodiment, the image data of the object to be identified can be the image data 30 of the cash register two-part invoice shown in Figures 3A and 3B, the three-part image data of the unified invoice shown in Figure 4A and 4B, and Figure 5A and 5B The image data 50 of the purchase exit or discount certificate shown, the image data 60 of an electronic computer unified invoice shown in Figures 6A and 6B, or the image data 70 of the Chunghwa Telecom bill shown in Figures 7A and 7B, but not Limited to the above.

With reference to FIGS. 1 and 2, please refer to step S12 to perform artificial intelligence recognition on the image data of multiple objects to be recognized to obtain artificial intelligence recognition results. In one embodiment, an artificial intelligence recognition device 22 is used to perform artificial intelligence recognition on the image data of multiple objects to be recognized to obtain artificial intelligence recognition results. In one embodiment, the artificial intelligence recognition device 22 is used to recognize image data to convert it into required item data, such as the item data of the aforementioned certificate.

With reference to FIG. 1 and FIG. 2, please refer to step S13 to determine the artificial intelligence recognition result to obtain a corresponding accuracy rate. In one embodiment, a judging device 23 is used to judge the artificial intelligence recognition result to obtain a corresponding accuracy rate. In one embodiment, the corresponding accuracy rate includes a corresponding success rate and a corresponding stability rate. In one embodiment, the judging device 23 of the visual artificial intelligence recognition system 20 of the present invention includes a comparing device 231 for comparing a set success value with the corresponding success rate, and comparing a set stable value with the corresponding success rate. Corresponds to the stability rate.

With reference to FIGS. 1 and 2, please refer to step S14 to determine whether to intervene a setting condition to the artificial intelligence recognition according to the corresponding accuracy rate, so as to perform the artificial intelligence recognition again. In one embodiment, according to the corresponding accuracy rate, an intervention device 24 is used to determine whether to intervene a set condition to the artificial intelligence identification device 22 to perform artificial intelligence identification again. In one embodiment, if the corresponding accuracy rate cannot meet the predetermined standard, for example, the artificial intelligence recognition result is broken or failed, and the recognition is not successful; or the stability is not good and the predetermined standard cannot be reached, then it is required The intervention device 24 is used to intervene the setting condition to the artificial intelligence identification device 22 to perform artificial intelligence identification again, so as to improve the corresponding accuracy.

With reference to FIGS. 1 and 2, please refer to step S15 to repeat the steps of artificial intelligence identification, determining the artificial intelligence identification result, and determining whether to intervene in a set condition, to complete the identification of multiple objects to be identified. In one embodiment, the artificial intelligence recognition device 22, the judgment device 23, and the intervention device 24 are used to perform artificial intelligence recognition, determine the result of artificial intelligence recognition, and decide whether to intervene in setting conditions, so as to improve the corresponding accuracy. And can successfully complete the identification of multiple objects to be identified.

In one embodiment, due to the various types and sizes of accounting voucher data, and the scattered distribution of voucher data items, if a large number of voucher image data are input to the artificial intelligence identification device 22, there is no Under the set conditions, although the artificial intelligence method is used for identification, it may cause the artificial intelligence identification to collapse, and the corresponding poor accuracy rate makes the reliability of the artificial intelligence identification result poor.

3A and 3B show schematic diagrams of image data of an object to be identified (two-part cash register invoice) according to the present invention. 4A and 4B show schematic diagrams of image data of an object to be identified (triple format of unified invoice) according to the present invention. 5A and 5B show schematic diagrams of the image data of an object to be identified (purchase exit or discount certificate) according to the present invention. 6A and 6B show schematic diagrams of image data of an object to be identified (electronic computer unified invoice) according to the present invention. 7A and 7B show schematic diagrams of image data of an object to be identified (Chunghwa Telecom bill) according to the present invention. With reference to FIGS. 1, 2 and 3A to 7B, the setting condition includes a first judgment condition, which is the aspect ratio of the image data of the object to be identified. According to the aspect ratio of the image data of the object to be identified, the aspect ratio of the image data of the object to be identified (cash register dual invoice) in Figures 3A and 3B is obvious compared to the aspect ratio of the image data of other objects to be identified It is different, so the image data identification of the object to be identified in Figures 3A and 3B can be classified as a cash register dual invoice.

In one embodiment, the first judgment condition further includes dots per inch (dpi, dots per inch) of the image data of the object to be identified. Using the dots per inch (dpi, dots per inch) of the image data of the object to be identified, the size of the image data of the object to be identified can also be identified to determine the type of image data of the object to be identified.

According to the above-mentioned first judgment condition, the image data of the object to be identified with obviously different aspect ratios can be initially identified. However, because many different certificates have the same size or the same aspect ratio, such as Figures 4A and The aspect ratio of the image data of the object to be identified in 4B (triple unified invoice) is the same as the image data of the object to be identified in Figures 5A and 5B (purchase exit or discount certificate), so the image cannot be identified using the first judgment condition The image data of 4A and Figure 5A. For example, the aspect ratio of the image data of the object to be identified (electronic computer unified invoice) in Figures 6A and 6B is the same as that of the image data of the object to be identified (Chunghwa Telecom bill) in Figures 7A and 7B, so the first judgment condition still cannot be identified The image data of Fig. 6A and Fig. 7A.

In one embodiment, if the first judgment condition and artificial intelligence recognition are used, and the corresponding accuracy rate is still not good, it is necessary to intervene in setting the condition, and the setting condition further includes a second judgment condition, which is to set the The image data of the object to be identified is a plurality of regions, and artificial intelligence identification is performed in at least one of the plurality of regions. In one embodiment, the second judgment condition is to set the image data of the object to be identified as a first area, a second area, and a third area. Referring to FIGS. 3A and 3B, the image data 30 of the object to be identified (cash register dual invoice) is set as a first area 31, a second area 32, and a third area 33. Perform artificial intelligence recognition in at least one of the first area 31, the second area 32, and the third area 33 to reduce the artificial intelligence recognition area, reduce the complexity of artificial intelligence recognition, and improve artificial intelligence recognition The corresponding accuracy rate.

In one embodiment, the setting condition includes inputting at least one predetermined image data. Referring to FIG. 3B, the predetermined image data 34 can be input to the artificial intelligence recognition device 22. Since the first area 31 in the double invoice of the cash register has the predetermined image data 34, the predetermined image data 34 can be identified in the first area 31 of the image data 30 of the object to be identified. It can be confirmed that the image data 30 of the object to be identified is the image data of the cash register dual invoice, and it can be identified and classified as the cash register dual invoice.

In one embodiment, referring to FIGS. 4A and 4B, the image data 40 of the object to be identified is set as a first area 41, a second area 42 and a third area 43. Perform artificial intelligence identification in at least one of the first area 41, the second area 42, and the third area 43, and input the predetermined image data (for example: unified invoice (triple type)) 44 to the manual Wisdom identification device 22. Since the first area 41 in the triple uniform invoice has the predetermined image data 44, if the predetermined image data 44 can be identified in the first area 41 of the image data 40 of the object to be identified, then Confirm that the image data 40 of the object to be identified is the image data of the triple unified invoice, and identify and classify it as the triple unified invoice.

In one embodiment, referring to FIGS. 5A and 5B, the image data 50 of the object to be identified is set as a first area 51, a second area 52, and a third area 53. Perform artificial intelligence identification in at least one of the first area 51, the second area 52, and the third area 53, and input the predetermined image data (for example, the purchase exit or discount certificate) 54 to the Artificial intelligence identification device 22. Since the first area 51 in the purchase exit or discount certificate has the predetermined image data 54, the predetermined image data 54 can be identified in the first area 51 of the image data 50 of the object to be identified , It can be confirmed that the image data 50 of the object to be identified is the image data of the purchase exit or discount certificate, and the identification is classified as the purchase exit or discount certificate.

In one embodiment, if the first judgment condition, the second judgment condition, and artificial intelligence recognition are used, and the corresponding accuracy is still not good, it is necessary to intervene in setting conditions, and the setting conditions include a third judgment condition , Set at least one sub-area in at least one of the plurality of areas, and perform artificial intelligence recognition in the at least one sub-area. 6A and 6B, the image data 60 of the object to be identified is set as a first area 61, a second area 62 and a third area 63, and two sub-areas 611 and 612 are set in the first area 61, At least one sub-area is recognized by artificial intelligence. And, input the predetermined image data (for example: an electronic computer) 64 to the artificial intelligence recognition device 22. Since the sub-area 611 of the first area 41 in the electronic computer uniform invoice has the predetermined image data 64, it can be identified in the sub-area 611 of the first area 61 of the image data 60 of the object to be identified The predetermined image data 64 can confirm that the image data 60 of the object to be identified is the image data of the electronic computer uniform invoice, and can be identified and classified as the electronic computer uniform invoice.

In one embodiment, if using the first judgment condition, the second judgment condition, the third judgment condition, and the artificial intelligence identification, if the corresponding accuracy is still not good, it is necessary to intervene in setting the condition again, the setting condition It includes a fourth judgment condition, which is related to the at least one sub-region, at least one side region is set, and artificial intelligence identification is performed in the at least one side region. 7A and 7B, the image data 70 of the object to be recognized is set as a first area 71, a second area 72 and a third area 73, and two sub-areas 711 and 712 are set in the first area 71, and In the sub-region 711, at least one side region 713 is set, and artificial intelligence identification is performed in the at least one side region 713. And, input the predetermined image data (for example: Chunghwa Telecom and its patterns) 74 to the artificial intelligence identification device 22. Since the side area 713 of the sub-area 711 of the first area 71 in the Chunghwa Telecom bill has the predetermined image data 74, if the image data 70 of the object to be identified is the sub-area 711 of the first area 71 In the border area 713, if the predetermined image data 74 can be identified, it can be confirmed that the image data 70 of the object to be identified is the image data of the Chunghwa Telecom bill, and it can be identified and classified as a Chunghwa Telecom bill.

Using the visual artificial intelligence recognition method and visual artificial intelligence recognition system 20 of the present invention can improve the accuracy of artificial intelligence recognition results, that is, the success rate of artificial intelligence recognition results and the stability rate of artificial intelligence recognition results. To improve the reliability of artificial intelligence identification results. Moreover, the use of the visual artificial intelligence recognition method and visual artificial intelligence recognition system 20 of the present invention can reduce the complexity of artificial intelligence recognition.

The above-mentioned embodiments only illustrate the principles and effects of the present invention, and do not limit the present invention. Therefore, those skilled in the art can modify and change the above-mentioned embodiments without departing from the spirit of the present invention. The scope of rights of the present invention should be listed in the scope of patent application described later.

20: Visual artificial intelligence recognition system 21: receiving device 22: Artificial intelligence identification device 23: Judging device 231: comparison device 24: Interventional device 30: Image data of the object to be identified 31: The first area 32: second area 33: The third area 34: Scheduled image data 40: Image data of the object to be identified 41: The first area 42: second area 43: The third area 44: Scheduled image data 50: Image data of the object to be identified 51: The first area 52: second area 53: The third area 54: Scheduled image data 60: Image data of the object to be identified 61: The first area 62: second area 63: The third area 64: Scheduled image data 611, 612: sub-area 70: Image data of the object to be identified 71: The first area 72: second area 73: The third area 74: Scheduled image data 711, 712: sub-area 713: side area S11~S15: steps

Fig. 1 shows a flowchart of the visual artificial intelligence identification method of the present invention.

Figure 2 shows a block diagram of the visual artificial intelligence recognition system of the present invention.

3A and 3B show schematic diagrams of image data of an object to be identified (two-part cash register invoice) according to the present invention.

4A and 4B show schematic diagrams of image data of an object to be identified (triple format of unified invoice) according to the present invention.

5A and 5B show schematic diagrams of the image data of an object to be identified (purchase exit or discount certificate) according to the present invention.

6A and 6B show schematic diagrams of image data of an object to be identified (electronic computer unified invoice) according to the present invention.

7A and 7B show schematic diagrams of image data of an object to be identified (Chunghwa Telecom bill) according to the present invention.

S11~S15: steps

Claims (18)

A visual artificial intelligence identification method includes: receiving image data of multiple objects to be identified; performing artificial intelligence identification on the image data of multiple objects to be identified to obtain an artificial intelligence identification result; judging the artificial intelligence identification result to obtain A corresponding accuracy rate; according to the corresponding accuracy rate, it is determined whether to intervene a setting condition to the artificial intelligence recognition to perform the artificial intelligence recognition again, wherein the setting condition includes a first judgment condition, which is the object to be recognized The aspect ratio of the image data; and repeat the steps of artificial intelligence identification, judging the artificial intelligence identification result, and determining whether to intervene in a setting condition to complete the identification of multiple objects to be identified. For example, the visual artificial intelligence identification method of claim 1, wherein the first judgment condition further includes dots per inch (dpi, dots per inch) of the image data of the object to be identified. For example, the visual artificial intelligence recognition method of claim 1, wherein the setting condition includes a second judgment condition, which is to set the image data of the object to be recognized as multiple regions, and perform at least one of the multiple regions Artificial intelligence identification. For example, the visual artificial intelligence recognition method of claim 3, wherein the second judgment condition is to set the image data of the object to be recognized as a first area, a second area, and a third area. For example, the visual artificial intelligence recognition method of claim 3, wherein the setting condition includes a third judgment condition, at least one sub-region is set in at least one of the plurality of regions, and the at least one sub-region is manually Wisdom identification. For example, the visual artificial intelligence recognition method of claim 5, wherein the setting condition includes a fourth judgment condition, at least one side area is set in the at least one sub-area, and artificial intelligence recognition is performed in the at least one side area. For example, in the visual artificial intelligence identification method of claim 1, the corresponding accuracy rate includes a corresponding success rate and a corresponding stability rate. For example, the visual artificial intelligence identification method of claim 7 further includes a comparison step for comparing a setting success value with the corresponding success rate, and comparing a setting stability value with the corresponding stability rate. For example, the visual artificial intelligence recognition method of claim 1, wherein the setting condition includes inputting at least one predetermined image data. A visual artificial intelligence identification system includes: a receiving device for receiving image data of a plurality of objects to be identified; an artificial intelligence identification device for performing artificial intelligence identification on the image data of a plurality of objects to be identified, Obtain the artificial intelligence identification result; a judging device for judging the artificial intelligence identification result to obtain a corresponding accuracy And an intervention device, according to the corresponding accuracy rate, determines whether to intervene a set condition to the artificial intelligence identification device to perform artificial intelligence identification again, wherein the set condition includes a first judgment condition, which is the to-be-identified The aspect ratio of the image data of the object. For example, the visual artificial intelligence recognition system of claim 10, wherein the first judgment condition further includes the points per inch of the image data of the object to be recognized. For example, the visual artificial intelligence recognition system of claim 10, wherein the setting condition includes a second judgment condition, which is to set the image data of the object to be recognized as multiple regions, and perform at least one of the multiple regions Artificial intelligence identification. For example, the visual artificial intelligence recognition system of claim 12, wherein the second judgment condition is to set the image data of the object to be recognized as a first area, a second area, and a third area. For example, the visual artificial intelligence recognition system of claim 12, wherein the setting condition includes a third judgment condition, at least one sub-region is set in at least one of the plurality of regions, and the at least one sub-region is manually Wisdom identification. For example, the visual artificial intelligence recognition system of claim 14, wherein the setting condition includes a fourth judgment condition, at least one side area is set in the at least one sub-region, and artificial intelligence recognition is performed in the at least one side area. For example, the visual artificial intelligence recognition system of claim 10, wherein the corresponding accuracy rate includes A corresponding success rate and a corresponding stability rate. For example, the visual artificial intelligence recognition system of claim 16 further includes a comparison device for comparing a setting success value with the corresponding success rate, and comparing a setting stability value with the corresponding stability rate. For example, the visual artificial intelligence recognition system of claim 10, wherein the setting condition includes inputting at least one predetermined image data.

Images