Image segmentation

Abstract

An image segmentation method using artificial intelligence to identify one or more vessels (or vehicles) in an image comprising a seascape (fig. 3). An image 202, which may be downscaled, is input into a first neural network to generate a first image segmentation result 204. Additionally, the image is input to a second neural network able to perform image segmentation on low contrast images e.g. infrared to generate a second segmentation result 206. Patches of the input image proximal to the horizon line can be sampled via a bi-geometric probability distribution and determining the horizon lines may require a third neural network. At least one of the segmentation results comprises a segmentation mask identifying one or more vehicles in the image. The output image (which may need upscaling) identifies the vehicles by combining the first and second segmentation results to create a mask 208. [Figure 2]