Guest article: What actually is Visual Search?

Users can take a picture of a product or object and upload it to a visual search engine through an app, a bot, website or any other tool and get exact information about it or a similarity and style-based recommendations.

How visual search works in detail

The search request image is sent to a visual search engine and as with any text search engine, the visual search engine performs a search within a specific dataset (the search index) for similar products or data and returns those ranked by similarity.

If the visual search index only consists of two products, a car and a dog for example, a search query with any car in the request image will return the car as the most probable result as there are no other similar options within the search index.

• Visual search is not the same as image classification. The latter tries to find the general class like chair, table or car for an object in an image. Visual search on the other hand, does not need a tag to classify an object correctly, it needs to find the most visually similar item in your data.
• Beyond that, visual search technology is not limited to finding visual similarities, but with supporting technologies like OCR (Optical Character Recognition) it can be used to quickly extract additional information from an image and enhance the results delivered by the visual search engine.

The rise of visual search is powered by the advancements in computer vision, especially in the field of deep learning and neural networks.

Additionally, due to the latest developments of in-memory databases and indexing technologies it is possible to search within millions, even billions of objects or images within milliseconds.

Technical and organisational implications of good search

The reason for limited text-based search functionalities is not the lack of mature technology but the overall challenge to provide and maintain enough and consistent metadata for digital assets. Manual tagging is often still a reality. Text-based search is also a tough nut to crack due to the ambiguity of natural language. Organisations must define common vocabularies and ensure their usage. However, external stakeholders and consumers will tend to abide by their own language nonetheless. Voice search does not ease the search complexity as spoken language must be also transcribed to text. Visual search helps to overcome these barriers.

A picture of a chair indicates the same in every language. Additionally, and contrary to common conception, visual search does not require images to be annotated with specific tags or classes. It is an often used but wrong approach. Modern, high precision visual search engines do not convert an image to a single class or tags that are understandable by a human. The whole advantage of having visual information would be lost, by trying to force visual search into a text-search pattern.

Visual Search is at its beginning, but it is moving fast.

Visual search is technologically advanced and bears a lot of business potential. Global brands have been quick to include visual search in their digital strategies. In some industries the use cases have been obvious from the start. Other industries might benefit from visual search as well, but do not have it at the top of their priority lists right now. In regard to visual search adoption, companies can still be divided into innovators and early adopters. As the technology is affordable and does not require large scale implementation projects, a mainstream adoption can be expected in the coming years.

Evaluating a Visual Search Vendor

Source: Nyris

Implementing visual search is a straightforward technology project if you work with a specialised visual search vendor. A test can be set up within a day without any integration requirement. This offers a quick and convenient way to check if visual search technology will provide benefits to your customers. Developing visual search in-house is also possible but expensive. It requires a team of dedicated deep learning experts to train and maintain your models and keep up to date with the current scientific developments. Additionally, backend engineers are required to create, maintain and deploy your visual search index and the AI models in scale. The costs can easily amount to a high 7-digit number to get the first solution running in production.

Here are six things you should consider when choosing a visual search vendor:

• Technological leadership - Visual Search is developing fast as a technology. Make sure that your technology partner integrates new algorithms and adapts their IT architecture constantly. A high involvement in R&D activities and regular updates are a good indicator. We strongly recommend testing different solutions before taking a final decision. To set up a test it is important to use the same test data and exactly the same request pictures for all vendors.

• Methodology - Implementing visual search and developing it as a valuable and frequently used feature can be easy if your technology vendor follows a comprehensive methodology. Get to know the requested efforts on your side in advance. Learn about the division of work in an ongoing cooperation.

• Data Privacy - Nowadays the privacy awareness of users is well developed and expectations for data security are high. With your visual search app you collect user-generated pictures and your technology partner reuses them for improving the applied machine learning algorithms. Make sure that you offer enough convincing data privacy options for your customers. Users should be able to easily choose manual settings by themselves. At least the following options should be available: deleting pictures immediately after metadata has been extracted and automatic deletion of user data such as geolocation, operating system and IP-address.

• Pricing model - Due to technological advancements, the costs of visual search have been significantly reduced in recent years. Most vendors charge per request and prices can range from 25 cents to a fraction of a cent per request. Make sure to compare the overall implementation price also in regards to included services and the overall costs when your search requests take off.

• Ensuring Adoption - Visual search can easily be implemented, but it takes more than that to turn it into a widely accepted feature that truly provides an added value. The functionality must be well integrated in the UI and companies have to provide incentives for users to try it out. It is a learning curve to recognize visual search as a useful enhancement. Companies need to promote this additional search feature. If you decide to extend your Customer Experience by visual search, consider not only technical aspects, but also ask yourself what your users can achieve with that feature and plan your launch communication accordingly. The usage of visual search will evolve. Companies focusing on gaining psychological insights and incorporating them in their offering will benefit from a competitive advantage.

 So… Choose nyris

Text-based search is a source of costly friction — especially for companies that manage products, component parts or other items that are essential to customer, supplier or employee success.

nyris is a visual search platform that gives people a more natural way to find what they are looking for. With nyris, companies of all industries realise financial value while improving how people find, discover and get things done. Up and running in minutes, only nyris’ technology is designed and engineered to handle any search challenge.

Based in Berlin and Dusseldorf, nyris serves leading companies across retail & e-commerce, industrial & manufacturing, media & entertainment and financial services. nyris serves customers active in 50 countries and representing every continent. Led by respected technology industry veterans, nyris investors include eCapital and Axel Springer.