10

Image analytics

What is it?

Image analytics is the process of extracting information, meaning and insights from images such as photographs, medical images or graphics. As a process it relies heavily on pattern recognition, digital geometry and signal processing.

In the past the only analysis that was possible on images was via the human eye or if computers were used they could only really assess the name of the image or any of the meta-tags that are stored against the image such as the date of creation, amendment, the owner of the image and the name it was given. For example, if you enter ‘pink elephant’ into an online search engine the search engine isn’t scrolling through the index to find pictures that match the description of a pink elephant, it is scrolling through the index to find meta-tags that match ‘pink’ and/or ‘elephant’. In other words when the person who uploaded the image uploaded the image they added descriptor key words to the image to help people find it. That said, just because the tag mentions pink elephant doesn’t mean that the image will relate in any way to pink elephants. As a result any analysis was basic and prone to errors.

Now image analytics is considerably more sophisticated. For example, image analytics is now being used on medical images of biopsies to help doctors identify cancer. In the case of photographs a digital photograph contains a lot more information than you might imagine; it will record when it was taken as well as where it was taken based on GPS coordinates embedded in the photo. All those additional properties can be analysed to extract more information above and beyond the actual image. Plus, one of the most exciting, and some would argue scary, developments in image analytics is face recognition.

When do I use it?

There are many applications for image analytics that could prove commercially relevant.

Face recognition analytics, for example, can automatically identify or verify a person from a digital image or video frame. This can be useful if you want to introduce an extra layer of security to your factory or premises. There could be an image of all your employees and image analytics would grant those people and only those people access to the premises. In addition, image analytics can potentially be used for marketing.

Facial recognition algorithms used to pick out facial features and analyse their relative position, size, shape, etc., or it would take a gallery of face images, normalise them and only save the distinct elements for facial recognition. These algorithms were therefore either geometric, which looks at distinguishing features, or photometric, which is a statistical approach that distils an image into values and compares the values with templates to eliminate variances. Advances in 3D face recognition technology mean that face recognition is now much more accurate.

Unsurprisingly, companies like Facebook are already ahead of the curve when it comes to face recognition because of the vast image data sets users willingly upload. Indeed Facebook researchers have reported that their DeepFace pattern recognition system is achieving near-human face recognition accuracy.

It is now possible, for example, for Facebook to recognise a face, compare it to previous photographs of that person and ‘decide’ if that person has put on weight. If they have, that data could then be sold to a weight loss company who would advertise on that users Facebook page.

What business questions is it helping me to answer?

Image analytics can help you secure your premises and help you to know more about your customers and what they are buying. Image analytics can help you to answer:

  • What and how many photographs contain our brand name?
  • Who are the customers that use our products?
  • How do we increase the security and improve access control?

How do I use it?

Obviously in order to use image analytics you need to have images to analyse. If you do have images or you already do have recorded video footage of your stores or premises then you could use that video footage as your source data. Alternatively, if you have access to other images such as medical images then you could use image analytics to extract insights. Of course there are tight laws around the use of data including images, so you will need to be mindful not to infringe on those laws.

Although image analytics can be incredibly useful in detecting patterns or anomalies in medical images and face recognition can be used for security and customer insight it’s probably not that appropriate for most businesses.

If you want to know more about the various image analytics techniques you can explore the links on page 40. Alternatively there are commercially available image analytics tools on the market that can help you.

Practical example

Casinos are currently using image analytics to identify high rollers for special treatment and presumably to identify people they want to keep out of their casinos, too. In Japan, grocery stores even use this technology to classify shoppers and blacklist serial complainers or shoplifters.

The biggest concern, especially around face recognition, is that it can be used without the person’s knowledge or consent. From a safe distance someone can covertly identify an individual by name which then connects to intimate details about that person such as home address, dating preferences, employment history and religious beliefs. In 2011 researchers at Carnegie Mellon reported that this was not a hypothetical risk when they used a face recognition app to identify some students on campus by name, linking them to their public Facebook profiles and, in some cases, to their Social Security numbers.

It is especially potent because of the internet. There is now so much image data online that businesses don’t even need to hold that data themselves for it to be useful. It is possible therefore to use image analytics to effectively scan the internet to gain insights and information about your customers and what offers or promotions they may respond to. In many ways image analytics represents a brave new world and it may only be a matter of time before legislation adds more controls and consumer safeguards.

Tips and traps

Like all analytics, image analytics is only really going to be useful if it helps to answer key strategic questions that you have as a business.

The biggest trap is privacy. Just because you can get your hands on images to analyse doesn’t necessarily mean you should, or that what you are doing is ethical. Make sure you have a specific reason for using image analytics that is morally defensible and where the outcome will deliver additional value to your customers.

Further reading and references

For more on image analytics see for example:

..................Content has been hidden....................

You can't read the all page of ebook, please click here login for view all page.
Reset
3.128.197.163