论文部分内容阅读
目的在分布式视频编码中,为了更加准确地描述相关噪声残差子带的变化特性,提出一种基于改进的模糊C均值(FCM)聚类的模型估计方法。方法本文算法针对每一解码子带选取不同的特征矢量;利用改进的模糊C均值进行聚类;采用阈值控制法求取相应的模型参数;然后用重建子带更新下一解码子带的特征矢量,直到一帧中所有子带解码完成。针对模糊C均值对初始聚类中心的敏感性,采用随机生成隶属度矩阵的方法来缓解聚类陷入局部最优的问题。结果从实验效果和算法复杂度角度考虑,将残差样本聚为8类。实验结果表明,本文聚类算法可以更加准确地模拟帧内不同区域的不同信道噪声特性,对于运动越剧烈的序列效果越好,相对于子带级拉普拉斯估计,平均增益达1 dB。结论提出了一种新的相关噪声估计方法,针对不同的子带选取不同的特征矢量,并重建更新。实验结果表明,本文算法能更好地描述相关噪声特性,获得系统性能的提高。
Aim In distributed video coding, in order to describe the variation characteristics of the correlated residual noise sub-band more accurately, a new model estimation method based on improved fuzzy C-means clustering (FCM) is proposed. Methods In this algorithm, different eigenvectors are selected for each decoding sub-band. Clustering is performed using improved fuzzy C-means. The threshold control method is used to obtain the corresponding model parameters. Then, the reconstructed sub-bands are used to update the eigenvectors Until all the sub-bands in one frame are decoded. Aiming at the sensitivity of fuzzy C-means to initial clustering centers, a method of randomly generating membership matrix is adopted to alleviate the problem of clustering falling into local optimum. Results From the perspective of experimental results and algorithm complexity, the residual samples were clustered into 8 categories. The experimental results show that the proposed clustering algorithm can simulate different channel noise characteristics in different regions of the frame more accurately, and the better the sequence is, the better the average gain is 1 dB compared with the subband level Laplacian. Conclusion A new correlation noise estimation method is proposed. Different eigenvectors are selected for different subbands and reconstructed. The experimental results show that the proposed algorithm can better describe the relevant noise characteristics and improve the system performance.