Skip to main content

Automatic image indexing with artificial intelligence (AI)

The company manages more than 300 campaigns a year for their various divisions in a central censhare Content Hub. All images for all internal and external uses are managed in the integrated Digital Asset Management (DAM). Not only do the images need to be quick and easy to find, they also need to be maintained with general and company-specific metadata and delivered to third-party systems via various interfaces. The previous system consisting of different feature lists had reached its limits and was not only cumbersome for users, but also no longer able to react quickly to adapted company structures.

Following a reconception, we integrated the required keywords directly into the information widget of the central content hub. The new system also supports an optimized search for structured and free search terms, the automated translation of all keywords and descriptions into the company’s four communication languages and the management of synonyms. This created the technical basis for implementing automated basic keywording with Amazon Rekognition. In addition to text recognition and facial analysis, this also enables the automatic recognition of objects and labels. In addition, company-specific features such as logos can be trained and identified using custom label recognition. All the required settings are integrated directly into the digital asset management system so that recognition takes place in the background without user intervention. And if more is recognized than required (e.g. an aircraft in the sky without reference to the image motif), the integrated approval process ensures that only content-relevant keywords are actually saved permanently.

The new keyword concept has many advantages: The search is fast and easy, the system performs much better than before and is intuitive for users without prior knowledge. The automatic basic keywording with artificial intelligence saves the company time and money, as all basic keywords are automatically recognized and assigned. Users only enter keywords that cannot be recognized automatically. And the future is already taken care of: by integrating all workflows and the configuration of the solution into digital asset management, the company can easily and quickly switch to a different model if it better meets its requirements.

Requirements

Swiss Post has long relied on a central content hub with censhare, which provides the company with digital asset management, print publishing and topic and campaign planning, among other things. The extensive image library is managed in Digital Asset Management with the help of various features. However, due to the volume of content and metadata, the previous system was not only slow, but also confusing and error-prone to maintain.

The new keywording concept was designed to optimize the handling and search of the images. Together with Swiss Post, we consolidated the various feature lists into a single keyword structure and integrated Amazon Rekognition into the Content Hub on this new technical basis in order to index the images quickly and automatically. The migration of the existing images in the system from features to keywords was also automated.

Implemented with
censhare

Use Case
Digital Asset Management

Implemented for
Die Post

Read Use Case
Download PDF

14000

keywords recorded

60

faster keywording

17000

images managed

A well thought-out DAM concept as basis for current and future requirements

In a first step, the existing keywording system was replaced by a new concept that meets the company’s increased requirements for flexibility, speed, performance and intuitive use. In this step, extended functions such as keyword hierarchies, synonyms and automated translation were also implemented.

Like all approval workflows, the configuration of the AI solution is now directly integrated into digital asset management and independent of the actual ML service. This means that users do not have to operate an additional system and the company can change the model if necessary without having to re-implement automatic keywording.

1. Simple keywording with fast search

Instead of many feature lists, there is now just a flat list of all the required keywords. The system suggests hits as soon as you start typing, making it very quick and intuitive for users to search for suitable keywords, even without knowing the hierarchy of a classic keyword tree. And the search for all images already tagged in the system is just as quick and easy.

2. Automatic basic keywording with approval process

After the upload and duplicate check, the DAM system sends new images to Amazon Rekognition, which is connected via an interface. There, the images are analyzed and returned with the keywords found. To prevent incorrect or irrelevant results, the results are sent to the relevant employees for final checking, approval and completion in a final workflow step.

3. Automated matching and translation

Each keyword is created as an asset in the system. When matching with Amazon Rekognition, the results are mapped to prevent duplicate entries. The keywords are then sent via an interface to the connected translation system, which takes care of the translations into 4 languages. All 4 languages and stored synonyms are then available to users in the system and the search.

4. Label recognition and custom model

Amazon Rekognition automatically recognizes a constantly growing number of supported objects and features. Important parameters for recognition, such as the confidential rate, can be adjusted directly in the DAM system if required. In addition, the system is trained for company-specific requirements and can therefore also recognize information that is not part of the standard.

5. Text recognition and face analysis

Text recognition captures individual words or entire sentences in images and saves them automatically. The final approval workflow prevents irrelevant entries and the settings for text recognition can also be adjusted directly in the DAM. Face analysis recognizes people’s emotions such as happiness or sadness as well as features such as beards or glasses and adds them as keywords.

”The new keyword system not only makes image searches faster and more intuitive, but also reduces the workload for our employees. They can now concentrate on their creative work while the AI takes over the routine tasks.

Michael EberleProduct Owner, Die Schweizerische Post

System Solution

censhare

censhare has fully integrated digital asset management, product information management and content management to help you to master your content. Companies like Allianz, Lands’ End, Dyson, Christie’s and hundreds more rely on censhare for on-brand, always up-to-date content, taking advantage of every opportunity to reach the right customers at the right time.

Learn more about censhare

censhare.com

The company

The Swiss Post is a public company that is 100% owned by the Swiss Confederation. In 2021, the company shipped around 1,8 billion letters and 202 million packages, transported 135 million passengers with PostBus and managed over CHF 111 billion in customer assets with PostFinance.

post.ch

Ready to go the path with us? We look forward to exchanging ideas with you.

Contact us