機器學習方法

專利類型

發明

專利國別 (專利申請國家)

中華民國

專利申請案號

109114083

專利證號

I 759731

專利獲證名稱

機器學習方法

專利所屬機關 (申請機關)

淡江大學學校財團法人淡江大學

獲證日期

2022/04/01

技術說明

企業員工與客戶進行會議時，與會者可能忘了對方的姓名及曾經合作的專案，即使參考過往的開會名單與照片，也面臨著照片人物與開會名單對應不上的問題。原因是照片中可能存在著未參加開會的朋友，或是照片中缺少開會的人。因此，在既無相片人臉標記，且會議名單中的姓名與相片中的人臉也不對稱的情況下，要自動辨識某個與會者的人臉變得困難。這樣的問題將造成企業無法透過以前合作的關係來拉近與客戶間的距離，洽談更多的合作案。由於上述動機，本專利設計 ＂基於人臉與人名不對稱資料配對之智慧化人臉即時識別系統＂，透過企業以往開會所留下的會議名單及會後的合照進行機器學習，讓系統能自動地、正確標記合照中人臉的姓名。 日後會議進行時，開會員工透過即時拍照，將照片傳回本系統，系統可辨識開會者的姓名，提醒開會參與人員之名單、熟悉與會人員的面貌、並使相片及開會人員能自動對應，以便拉近客戶關係。本專利主要先採用機器學習非監督式學習結合演算法的技術，標記照片中的人臉後，再採用監督式學習辨識人臉，對開會人員的相片進行辨識與標記姓名，並且關聯與會人員之相關資訊及檢視是否有人員被遺漏。 When a company employee meets with customers, it is quite common for the employee to have forgotten the name of the other party and projects that the two companies have cooperated with. Even if they refer to the list and photos of previous meetings, they also have the problem that the persons in the photo can't match with the meeting attendee list. The possible reason is that there may be people in the photo who did not attend the meeting, or those who attended the meeting are not in the photo. No tagging of each person in the photo along with the attendee list and photo matching problem makes it difficult to recognize the face of a particular pmiicipant. This challenge may prevent companies from getting closer to customers and unable to negotiate more cooperation cases. Due to the above motives, this invention patent design " Intelligent Real-time Face Recognition System based on Asymmetric Face Pictures and Name List. 11. Given the previous meeting lists and photos, this patent proposes a new approach which applies machine learning techniques, including deep learning mechanisms, aiming to tag the name of each face in the photo automatically, immediately and correctly. With this system the meeting staff can send meeting photos to the system, the system can identify the name of the meeting attendee, remind the user of the list, familiarize the face of the meeting participants, match the photos and meeting attendee automatically and hence enhance customer relationships. This patent first adopts unsupervised learning to tag the face in the photo, then uses supervised learning to identify the face as well as tag each face of the meeting attendee in the photo. At the same time, it can provide relevant information about the meeting participants and check if any people have been omitted.

備註

連絡單位 (專責單位/部門名稱)

研究發展處

連絡電話

26215656分機2561

網址