ShoppingTotal: A Mobile Application Utilizing Assisted Rekognition Algorithm for Intelligent Price Detection from Shelf Label Images

Abstract

ShoppingTotal is a mobile application for monitoring the shopping budget through shelf label images. Using the ShoppingTotal application, shoppers capture the shelf label image of the product to obtain the product information and view the total amount of the current shopping and the history of the previous shopping lists. For the ShoppingTotal application, the Assisted Rekognition algorithm is developed based on Amazon Rekognition’s text detection service for extracting product information from label images. The FourGroceries dataset is collected for evaluating the performance of the Assisted Rekognition algorithm over original, single-filtered, and multifiltered images based on the image filters under the categories of sharpness, blurriness, brightness, temperature, and color. According to experiments on the FourGroceries dataset and the Amazon Rekognition service, the average price detection confidence results are 76.49% with the Assisted Rekognition algorithm and 20.94% without the Assisted Rekognition algorithm. The Assisted Rekognition algorithm’s performance is found to be better on filtered images than on original images, with 89.25% price detection confidence. By applying appropriate single or multiple image filters on the FourGroceries dataset, the Assisted Rekognition algorithm achieves extracting the correct price values from all experimental dataset images.