Underwater laser image segmentation method based on adaptive pulse coupled neural networks

Range gated underwater laser imaging technology, which has broad application prospects in oceanic research, deep sea exploration and under water operation field, is one of the most effective methods to decrease the backward scattering effect of water medium. However, the special features of underwater laser images, such as speckle noise and non-uniform illumination, bring great difficulty for image segmentation. By analyzing the formation principle of speckle noise, an effective underwater laser image segmentation method is proposed. On the basis of noise response and intensity distribution, the proposed method determines the certain key parameters of neurons adaptively, while suppesses the behavior of neurons located in speckle noise. A gradient descent algorithm based on criterion of maximum two-dimensional Renyi entropy is applied to determine the dynamic threshold of neurons. Experimental results demonstrate that the method is significantly superior to Normalized Cut, fuzzy C means, mean shift and watershed methods, while the consumed time of this method is about one-fifth of conventional pulse coupled neural networks. ©, 2015, Chinese Optical Society. All right reserved.