Modal MRF (Markov Random Field) based underwater forward-looking sonar image segmentation method

Abstract

The invention relates to a modal MRF (Markov Random Field) based underwater forward-looking sonar image segmentation method. The core of the invention is that modeling and segmentation are performed on an image by using a modal Potts network. The method specifically comprises the steps of initializing a Potts state; estimating class conditional parameters by using an EM algorithm; updating each Potts unit according to heating bath power, wherein an internal field of the Potts unit meets Markov dependency, and an external field of the Potts unit is a logarithm of the class conditional probability of pixels; updating interaction parameters between the Potts units and interaction parameters between the Potts units and the external field; and judging whether to exit or not according to the convergence degree of the parameters. According to the invention, image segmentation is performed by adopting the heating bath power based Potts network, thereby conforming to a cognitive mechanism of the brain. The method can almost perfectly preserve edge information of a target while removing plaque noise.

Description

technical field [0001] The invention belongs to the field of underwater sonar image processing, in particular to a forward-looking sonar image segmentation method based on modal Markov random fields. Background technique [0002] Underwater robots are carriers for exploring the unknown underwater environment and performing specific underwater tasks. Since the photoelectric signal has a very limited operating distance in seawater, sonar has become a basic and necessary tool for underwater robots to perceive the underwater environment. Sonar images play a very important role in marine environment monitoring, seabed resource exploration, underwater acoustic intelligence collection, and assisting underwater engineering operations. Among them, forward-looking sonar is the preferred tool for close-range surveys and operations. [0003] However, the forward-looking sonar image contains a large amount of speckle noise due to the interference of water body reflection and sound wave...

AI Technical Summary

Problems solved by technology

However, there are the following disadvantages: (1) The ability of the MRF model to deal with plaque noise is limited, and in most cases it is necessary to perform morphological post-processing to remove noise with a higher prior probability

(2) Although the EMMF algorithm has a good smoothing effect, there is a phenomenon of over-smoothing, which will lose a lot of contour information

Images