本文提出了一種新的基于極大似然法的椒鹽噪聲濾波算法。在傳統(tǒng)BP 算法中引入了極大似然估計(jì)，在訓(xùn)練樣本時(shí)能夠在考慮網(wǎng)絡(luò)逼近行為的同時(shí)對(duì)噪聲分布進(jìn)行估計(jì)。而且
針對(duì)椒鹽噪聲模型構(gòu)造了新的魯棒誤差函數(shù)，從而使算法本身的抗干擾性增強(qiáng)。實(shí)驗(yàn)結(jié)果表明了該算法與傳統(tǒng)BP 算法相比具有更好的濾波性能。
關(guān)鍵詞 BP 算法 椒鹽噪聲 極大似然法 魯棒誤差函數(shù)
Abstract One kind of new salt and pepper noise filter algorithm based on the maximum
likehood method was proposed in this article.The maximum likelihood was introduced into the
traditional BP algorithm for the sake of estimating the noise distribution during the network
approaching of the training sample.And in order to enhance algorithm BP’s Anti-noise property,a new robust error function was accordingly defined in view of the salt & pepper noise.The experimental results indicated our approach has the better filter performance compares with the traditional BP algorithm.
Keyword algorithm BP salt&pepper noise maximum likelihood robust energy function