Real-time image segmentation software

The partitioning of digital images into multiple segments consisting of sets of pixels (superpixels) is a crucial technique in computer vision. Applications can be found, for example, in object recognition and 3D reconstruction of environments. However, state-of-the-art segmentation methods require significant computational resources, which preclude their use in real-time systems. It is possible to reduce this dependency on the availability of computational infrastructure by reducing the number of processes, however, this results in reduced performance.

Oxford researchers have developed gSLICr, image segmentation software that is optimised to provide real-time applications on a modern GPU without loss of performance.

Image segmentation

Image segmentation methods are widely used to simplify the representation of a digital image in order to make it easier to analyse. This can be achieved by partitioning the image into sets of pixels known as superpixels, a method that is becoming increasingly popular. Image segmentation has a number of important applications in computer vision, including object recognition and 3D reconstruction of scenes.

Simple Linear Iterative Clustering (SLIC)

The majority of image segmentation methods currently require a significant amount of processing power to implement, rendering them unsuitable for real-time applications. The development of the Simple Linear Iterative Clustering (SLIC) method provided a simple, efficient, and high-performance algorithm which increased practicality over existing methods. However, previous implementations of SLIC could not provide sufficiently high-speed processing without significant loss of performance for real-time applications.

gSLICr – real-time image segmentation software

Oxford researchers have developed gSLICr, image segmentation software that offers real-time segmentation without loss of performance. gSLICr is the fastest GPU implementation of the SLIC algorithm, improving the speed of superpixel segmentation 100 fold over the current state-of-the-art. Use of the gSLICr software package will enable high-quality superpixel image segmentation to find real-time applications in computer vision. Isis Innovation is seeking industrial partners who wish to explore the use of gSLICr for commercial applications.


Request more information
about this technology

Ready to get in touch?

Contact Us
© Oxford University Innovation