平價自助餐之自動化影像辨識計價系統

Automated Pricing System for Parity Cafeteria based on Image Recognition

江昱賢(Yu-Xian Jiang)

指導教授 ： 林宏達

朝陽科技大學/理工學院/工業工程與管理系/碩士(2020年)

摘要

平價自助餐因食物種類多、方便、且經濟實惠為學生、上班族與外食族所青睞，自助餐的計價方式幾乎多採人工目視搭配人員心算而成或是採秤重方式計價。大部分自助餐沒有標示每樣菜色的價錢，對消費者來說權益沒有受到保障，也因此造成許多紛爭，媒體常登載因份量多寡或種類差異之認定、單價弄錯、加總計算錯誤等因素而造成紛爭。本研究針對平價自助餐之計價方式、為消弭因人為因素所造成之紛爭，並且加快計價時間等因素考量進行探討。本研究目的要發展一套自助餐自動化視覺計價系統，希望透過此系統能正確判斷食物種類與數量進而計算餐點價錢，達到量價相符以提高整體服務品質。
本研究的研究方法分為兩部分，第一部分為以傳統物件分割為基礎的影像辨識法，在傳統物件分割法中一開始需要分割出前景與背景，再將前景分割為許多個別物件，再針對個別物件編號並進行後續處理。第二部分為以現代影像分格為基礎的影像辨識法，在現代影像分格法中不分割前景與背影，直接將影像分格為均勻大小的格子區塊，再針對各個格子區塊進行後續處理。本研究使用310張自助餐樣本影像，實驗結果顯示以本研究之第二部分之方法可以有效辨識餐盤中的食物類別與份量。本研究之食物正確分類率(1-γ)為93%、分格區塊正確分類率(CR)為92%。

關鍵字

食物辨識 ； 平價自助餐 ； 均值偏移濾波 ； 分水嶺分割法 ； 類神經網路 ； 自動計價系統

並列摘要

The pricing of cafeteria almost take human visual collocation staff mental arithmetic or adopt weighing by scale. Cafeteria does not indicate the price of each dish, for consumers, there is no guarantee. As a result, many disputes have taken place, the news often report due to the amount of weight or wrong identification of food types, wrong unit price of food, the erroneous total amount and other factors caused disputes. In order to eliminate disputes caused by human error and accelerate the pricing time. This paper aims to develop a set of Automated Pricing System for Parity Cafeteria based on Image Recognition. We hope that through this system, we can correctly identify the food type and quantity of food value and then calculate the correct price of the meal, so as to achieve the consistent price and quantity to improve the overall service quality.
The research method of this research is divided into two parts. The first part is the image recognition method based on traditional object segmentation., the foreground and background need to be segmented at first, and then the foreground is divided into many individual objects. Individual object numbers and subsequent processing. The second part is an image recognition method based on modern image division. In the modern image grid method, the foreground and the background are not divided, and the image is directly divided into grid blocks of uniform size, and the follow-up is performed for each grid block. This study used 300 images of cafeteria samples. The experimental results show that the method of the second part of this study can effectively identify the type and amount of food in the plate. The correct classification rate (1-γ) of food in this study was 89%, and the correct classification rate (CR) of sub-blocks was 92%.

並列關鍵字

Optical vision system ； Parity cafeteria ； Mean shift filter ； Watershed segmentation ； Neural network models ； Automatic pricing valuation

參考文獻

[1] Abuzneid, M., Mahmood, A., “Enhanced Human Face Recognition Using LBPH Descriptor Multi-KNN and Back-Propagation Neural Network,” IEEE Long Island Systems, Applications and Technology Conference (LISAT), vol. 6, pp. 20641-20651 (2018).

[2] Ai, L., Xiong, J., “Temporal–spatial mean-shift clustering analysis to improve functional MRI activation detection,” Magnetic Resonance Imaging, Vol. 34, pp. 1283-1291 (2016).

[3] Anthimopoulos, M., Dehais, J., Diem, P., and Mougiakakou, S., “Segmentation and Recognition of Multi-Food Meal Images for Carbohydrate Counting,” IEEE International Conference on BioInformatics and BioEngineering, (2013).

[4] Apapajeeta, J., Mahakud, S., Nanda, P. K., and Das, N., “Variable Variance Adaptive Mean-Shift and possibilistic fuzzy C-means based recursive framework for brain MR image segmentation,” Expert Systems With Applications, Vol. 92, pp. 317-333 (2018).

[5] Arnab, C., Arpita, D., and Mahua, B., “Mammographic Image Segmentation by Marker Controlled Watershed Algorithm”, IEEE International Conference on Bioinformatics and Biomedicine (BIBM), (2017)