Since the inception of modern search engines, when users had a certain search query in mind, they would type the information into a search engine’s query box. However, if your inquiries cannot be translated into written words, how can you do a search?
For example, there may be a bag that a celebrity carried on TV or in a magazine, or a type of food whose name you do not know. The same goes for any general purpose products. When the details are missing or cannot be specified, it is not easy to get information from search engines.
In order to remove such search inconveniences, Add Quality Inc. has been developing an Artificial Intelligence (AI) search engine called “Ingram”. Ingram allows people to conduct searches by submitting photos as the search query. On the app, photos can be taken, analyzed, and searched to identify the product and locate similar ones. Additionally, it is capable of suggesting the way to dress based on the accumulated fashion-search history data in your account.
Ingram is an AI system that can perceive and process images in a similar way as human beings. It can learn any abstract concepts such as “bags” or “pasta”. When it takes the visual information, it would extract the semantic information of the objects from its database. For example, receiving the image of “spaghetti bolognese” it would associate it with other visual and textual information like “carbonara”, “ramen noodle” “Thai fried noodle”, etc. Out of all those conceptual food groups, the semantically nearest ones are to be selected as the search results. It would learn to define what “bolognese” is after the analysis process and enable itself to conclude the correct answer naturally.
How does it work?
Below is the image from the official Ingram page. If you press the camera button located top center of the mobile screen, it will take in the image to acknowledge and analyze the products in the frame. In this specific photo, the application will recognize the bag, the gloves, the coat and the sunglasses and proceed to list up the similar products for each item.
Soichi Matsuda, the CEO, says, “Conventional machines could not manage any undefined, obscure tasks. Only by acquiring this human ability, it would search and specify a variety of objects in a wide range.” As to clothing coordinations, learning that white jackets match with blue pants would enable it to suggest feasible clothing coordination patterns.
The accuracy of AI is propotional to the amount and quality of data it has collected. Matsuda considered that all the collectable online data as still not useful enough to materialize more detailed text and visual data that any searchers would want. Add Quality has engaged Nissen, a mail-order company affiliated with Seven and i Holdings Co., Ltd., to provide a database composing of fashion-related visual and text data. In addition, Seven and i Holdings group’s exceptional business range made it possible to provide a substantial product database.
What does the future have in store?
The company is now considering adding the following services to Ingram:
- Recipe provision services – giving recipes based on the images of leftover ingredients.
- Restaurant guide services – suggesting a menu based on the photos of meals accumulated on one’s mobile.
- Health care services – assessing symptoms by photos and recommending whether or not you should see a doctor.
Matsuda CEO says, “There currently is no standard set for AI marketing. We want to be the one to set it.” The search engine has not been launched as of July 2015, but once it does, we will review its impact on other search engines in Japan.