Sign in Subscribe
Machine learning

Patent Granted for Neural Network Based Image Processing Method

Strategic Insights

— 1 min read

A significant breakthrough in the field of artificial intelligence has been achieved with the grant of a patent for a neural network based image processing method. The patent, filed by Tom Tabak of Tel Aviv, Israel, and assigned to Autobrains Technologies Ltd., describes a novel approach to processing two-dimensional images using machine learning techniques. This innovative method has the potential to revolutionize various applications, including computer vision, object detection, and image classification.

Key Takeaways:

  • The patented method involves obtaining multiple two-dimensional segments of a sensed information unit, converting each segment to a segment vector, and generating intermediate results by multiplying each vector by a learnable embedding matrix.
  • The intermediate results are then concatenated with associated segment identifiers to provide a sensed information unit result, which is fed to a second layer of a neural network for content and location aware processing.
  • The method can be executed by a non-transitory computer readable medium and can be trained using supervised or unsupervised training processes.
  • The invention relates to neural network based image processing and provides a solution to the need for machine learning processes that can benefit from encoders developed in the NLP domain.
  • The patented method can be applied to various applications, including computer vision, object detection, and image classification.

Statistics:

  • The patent, U.S. Patent Number 12322110, was filed on October 11, 2022, and published online on June 3, 2025.
  • The patent has 12 claims, including methods for neural network based image processing, non-transitory computer readable mediums for storing instructions, and image processors for executing the methods.
  • The neural network used in the patented method can be trained using supervised or unsupervised training processes.

Sources:

  • Tabak, Tom. Generation of concepts for lane and road boundary prediction. U.S. Patent Number 12322110 (filed October 11, 2022).

Read more