计算机工程 ›› 2018, Vol. 44 ›› Issue (5): 252-255.doi: 10.19678/j.issn.1000-3428.0046947

• 多媒体技术及应用 • 上一篇    下一篇

基于频率敏感三维自组织映射的立体视频视差估计算法

任云,程福林,黎洪松   

  1. 桂林电子科技大学 信息与通信学院,广西 桂林 541004
  • 收稿日期:2017-04-25 出版日期:2018-05-15 发布日期:2018-05-15
  • 作者简介:任云(1992—),女,硕士研究生,主研方向为视频通信、神经网络算法;程福林,硕士研究生;黎洪松,教授、博士。
  • 基金项目:
    国家自然科学基金(61261035)。

Stereo Video Parallax Estimation Algorithm Based on Frequency Sensitive Three-dimensional Self-organizing Mapping

REN Yun,CHENG Fulin,LI Hongsong   

  1. School of Information and Communication,Guilin University of Electronic Technology,Guilin,Guangxi 541004,China
  • Received:2017-04-25 Online:2018-05-15 Published:2018-05-15

摘要: 传统基于块的立体视频视差估计算法计算复杂、效率较低,难于以硬件实现。为此,提出一种基于三维自组织映射的视差模式识别算法。根据图像块均值的大小,为视差序列建立高、底亮度模式库,输入块根据均值大小在不同的库中匹配模式块,从而实现视差估计。在模式库训练时引入频率敏感方法,以提高矢量利用率,减少死神经元,降低获胜矢量响应频率。实验结果表明,与基于块的视差估计算法相比,该算法的视差预测图像的平均峰值信噪比提高了1.78 dB。

关键词: 视差估计, 自组织映射算法, 聚类算法, 视差模式识别, 立体视频编码

Abstract: Traditional stereo video parallax estimation algorithm based on blocks is difficult to be implemented in hardware,because of its complexity and low efficiency.To solve this problem,parallax pattern recognition algorithm based on three-dimensional self-organizing mapping is proposed.According to the mean values of image blocks,establish pattern library with high brightness and pattern library with low brightness for the parallax sequence.Input blocks can search their matching pattern blocks in different libraries according to mean values.In the library training,the frequency sensitive method is introduced to improve the utilization of vectors,reduce the dead neurons,and decrease the response frequency of the winning vectors.In the experiments,the average Peak Signal-to-Noise Ratio(PSNR) of the algorithm is 1.78 dB,and that results are higher than the traditional parallax estimation algorithm.

Key words: parallax estimation, self-organizing mapping algorithm, clustering algorithm, parallax pattern recognition, stereo video coding

中图分类号: