Finding the Ollie in the Haystack: Exploring how AI is Revolutionizing Visual Search

You ask, “What’s an ollie?” Hang on to that question and read on.  

Wouldn’t it be nice to live in a world where photographers, videographers, editors, art directors, and people with similar roles could visually describe what they are looking for without considering the keywords and metadata used to describe the desired media?  

 It’s a puzzle we all live with, and it’s especially true and painful for video.  

 What if a user could visually describe a scene, and it was instantly queued up to the right frame, right in front of them? Think of all the B-roll videos that remain undiscovered.  

 What if metadata was no longer needed to search for media? After all, the expense of creating metadata for all your content can be prohibitive. 

 Let’s explore how AI is enabling users today to search how they think and not be constrained by unwieldy keyword lists and deep taxonomies.  

Digital asset management and metadata search 

It seems like digital asset management (DAM) systems have been a staple forever. Their inception dates to the early 1990s, nearly three decades ago. DAMs were born out of necessity as high-quality, low-cost digital scanners, and early digital cameras emerged.  

That seems like a lifetime ago, right?  

Fast forward to 2024. We continue to search for visual objects using textual search terms, hoping those same terms (“metadata”) happen to be attached to the visual asset by someone or some machine. 

Designers, researchers, producers, scholars, and others use text-based search queries to find visual content for diverse media channels every day. 

Relying on a keyword-based search (metadata) is like reaching into a bag of unknown words not knowing what you will pull out. In short, it’s inefficient, and you’re likely missing the best content.

The exponential growth in content is making manual content tagging economically unfeasible.

Even with the introduction of machine tagging solutions, they frequently fall short in providing visual context. They also tend to add irrelevant metadata, simply introducing noise to your search results. 

By no fault of their own, DAM managers, archivists, and researchers struggle to keep pace by adding necessary metadata to make the latest content submissions discoverable.  

 DAM users invest valuable time in the content discovery process, with many grappling to construct intricate Boolean queries and use a checkbox list of filters to pinpoint the content they are looking for.  

Frustrated by the complexity and poor search results, users often give up and defer the task to the research team or a seasoned user, adding additional work to an already burdened team. 

 Sound familiar? The struggle for discovery is real.  

A vast multimedia database becomes searchable with AI 

In my prior role at the Associated Press (AP), we embarked on redesigning the AP content delivery platform to bring customers a best-in-class multimedia experience.  

As sharp as the design and related tools were, our customers’ number one complaint centered around their difficulty searching for content.  

It doesn’t matter how good the user experience is, how vast the taxonomy, or how many search filters there are. Those features are irrelevant if you can’t find what you are looking for.  

The Associated Press’ vast archive contains tens of millions of images and millions of minutes of video. It supports thousands of customers around the world, performing millions of searches that span from rich historical content to breaking news.  

Today’s users are busy: deadlines are tight, and the demand for content has no bounds.

Every user requires accurate, fast search results. 

Each person thinks differently. Teaching or expecting your customer to understand a deep taxonomy or memorize a list of keywords used to describe visual assets is impractical.  

Equally as important, poor search results will negatively impact your teams and business. Due to modern search engines, there is a whole generation of digital native users who have grown up with cellphones in hand and the world their fingertips. They expect instant access to vast amounts of information and a frictionless discovery experience.  

It’s time to give your users a natural and intuitive visual search experience to find your content.  

Thanks to advancements in deep learning, that’s now possible.  

Instantly find content using the latest in AI-powered visual search 

Canto’s VisualSearch Pro is helping AP’s customers to quickly unlock content hidden amongst tens of millions of assets. This new AI-based search tool visually analyzes image and video content without dependence on metadata or automated tagging.    

Let’s go look for that “ollie in the haystack.”  

A user is working on a project and needs an image of: a gen z er doing an ollie at sunset.  

You ask, “What’s an ollie?” And then you ask, “Would someone really tag content with the words gen z er, ollie?” For non-skateboarders, an ollie is a skateboarding trick where a skater catapults into the air with their board without using their hands. 

Due to the nature of this search, it’s highly doubtful that a human or machine would properly tag these assets to discover content using these search terms.   

Canto’s VisualSearch Pro easily and quickly finds the Gen Z-er doing an ollie at sunset out of tens of millions of images.  

At AP, VisualSearch Pro allows users to search how they think using natural language. For example, a user might search for a video clip showing President Kennedy smiling and sailing with Jackie.  

VisualSearch Pro searches millions of minutes of video, returning a clip showing the user precisely what they searched for. Not only can it find the clip, but as a bonus, it takes the user to the precise moment at 16:15 in a 24-minute clip of them sailing.  

Here’s one more search that few users would ever try because they have been conditioned to use two or three words at most in their search: blurred motion shot of a skier wearing red skiing through trees.  

Is this the end of metadata? 

Does the advent of VisualSearch Pro mean the end of metadata? Most certainly not.  

Canto’s industry-leading AI-powered visual search marks a transformational moment in the evolution of search technology. The dependency on adding descriptive metadata and the need to teach users how to navigate complex libraries is becoming obsolete.  

Don’t take my word for it. Try it out for yourself. 

Go to the AP Newsroom , click the Enable AI button, and take it for a spin.