Panasonic Refines Image Recognition Technology with AI

Panasonic has announced a new AI-driven image recognition algorithm that is capable not only of identifying and tracking various subjects but also of dividing them into subcategories based on how they appear in the image.

Panasonic X AI is capable of further identifying an object even it’s appearance is impacted by conditions that traditional object identification has trouble with.

The announcement goes on to say that though traditional AI methods have difficulty with this multi-modal distribution technique, Panasonic has made it work by using a benchmark dataset* to perform highly accurate image classification.

Image Credit – Panasonic HD

The process was developed in response to the expanding use of AI for subject recognition and applications within a variety of visual mediums.

Image Credit – Panasonic HD

The new technique, known as image recognition AI, can pick up a subject even when its appearance is affected by lighting, orientation, weather, or background.

Moreover, the technique known as “multimodal distribution,” can further divide a subject into various subcategories such as the breed of an animal, type of vehicle, or even various appearances such as color.

“AI often has trouble successfully recognizing such objects as being in the same category, resulting in a decrease in recognition accuracy,” states the Panasonic announcement.

“Therefore, our company has focused on taking advantage of differences in appearance and developed a new classification algorithm that captures the diversity of images using a two-dimensional orthonormal matrix.”

The technique can also further subdivide the same subject-object based on various activities. So the AI can identify a bird, for instance, that is sitting on a branch, and then pick up that same bird in flight or even landing.

“In order to continuously capture the distribution of features, we expanded the weight vector of the classification model, which has traditionally only been a one-dimensional vector, to a two-dimensional orthonormal matrix,” the announcement continues. “This allows each element of the weight matrix to represent a variation of the image (differing background colors, object orientation, etc.).”

How long it will take to find its way into various cameras is anyone’s guess, but Panasonic has plans to further develop the object recognition scheme and then once matured, deploy the scheme industry-wide. The company will then promote the research and development of the technology to accelerate its social implementation while also focusing on training top AI experts in how to further develop and deploy it.

The company will be presenting its research at the plenary IEEE/CVF Winter Conference on Applications of Computer Vision (WACV) conference in Hawaii, which runs from January 4 to January 8, 2024. The white paper outlining details of the technique can be downloaded from the WACV website.

[source: Panasonic HD]

Disclaimer: As an Amazon Associate partner and participant in B&H and Adorama Affiliate programmes, we earn a small comission from each purchase made through the affiliate links listed above at no additional cost to you.

Claim your copy of DAVINCI RESOLVE - SIMPLIFIED COURSE with 50% off! Get Instant Access!

Leave a Reply

This site uses Akismet to reduce spam. Learn how your comment data is processed.