Speckle Noise Removal By Genetic Neuro-Fuzzy System In Sonography Images

A new filter was designed and approved for speckle noise removal in sonography images. In this filter, a new idea is used by using neural network learning, fuzzy information and genetic algorithm optimization. The multi-layer perceptron neural network with binary weights is used in this filter. The neighborhood window of each pixel is used as input statistical features to estimate the noise level. Then it is fuzzificated and justified by simple fuzzy rules. The membership function width and network weights are optimized by on-line genetic algorithm. The on-line algorithm contains one individual, defined as a queen. In this algorithm, the next generation is created by using only the mutation operator. The performance of this filter was compared with the other speckle noise reduction techniques such as the median and homomorphic Wiener filters. Indeed, our proposed method is able to effectively remove speckle noises while preserving the quality of fine details in the image data better than the other methods. In this system, two classic and on-line GAs are used. The classic algorithm includes 50 strings. The results showed that both of the algorithms are the same in terms of noise reduction but the classic one is slower than the other one.