【摘要】 根据等失真(Equidistortion)理论提出了一种基于改进的自组织特征映射(SOFM)神经网络的矢量量化方法,该算法将失真敏感机制引入神经网络的竞争学习过程。通过调整码字的部分失真来指导神经网络的学习,以使得所设计的码书平均失真最小。同时把矢量量化应用于图像的小波变换域,根据图像小波变换高频系数的空间分布特点来组织码书,从而进一步提高码书的质量和适应性。通过实验对算法的性能进行了分析,证明了算法的有效性。更多还原

【Abstract】 A codebook designing algorithm of vector quantization based on modified self-organizing feature maps(SOFM)was proposed.Distortion sensitive was inducted into competitive learning of neural network.Learning was instructed by modifying the distortion of every codeword,therefore,codebook could achieve the minimal distortion.Furthermore,this algorithm is applicable in wavelet transform of origin image.And,characteristic of image frequency distribution can be utilized sufficiently.From do it,the performance of codebook can be improved.The experiment indicates the effectiveness of the algorithm.更多还原