基于模糊神经网络的脉冲噪声滤波器

摘要

针对一般模糊神经网络结构复杂、不利于硬件实现的问题，提出了一种基于Sugeno型模糊神经网络的新型脉冲噪声滤波器.该滤波器采用神经网络的结构设计，有利于噪声模式的检测，其内含于神经网络中的模糊推理机制不仅能够有效地滤除脉冲噪声，而且又不破坏图象的细节.该滤波器还采用能够获得全局解的遗传算法来对网络参数进行调整.初步研究表明，该模糊神经滤波器在滤除景物图象中的脉冲噪声方面，优于标准中值滤波器.%A new filter based on a Fuzzy Neural Network(FNN) of Sugeno type is presented for images corrupted by impulse noise.Impulse noise results in the quality decline of image and can be reduced by nonlinear image filters, such as FNN image filter. FNN image filter does better than other kind fo filters when judged of subjective vision quality because its way of working is even close to that of mankind's eyes. The network structure of filter is good at ditecting different patterns of noisy pixel while the fuzzy mechanism embedded in the network can remove impulse and keep details and textures. Sugeno type NN have simple structure and other merit, which makes it suit for constructing FNN filter. A learning method based on the genetic algorithm is adopted to adjust the network parameters from a set of training data. The preliminary experimental result shows that the Fuzzy Neural Network filter performs better than the Median Filter when used to cancel impulse noise from scene image.