屏幕内容索引图的马尔可夫预测算法

发布时间：2018-04-04 10:34

  本文选题：图像编码　切入点：屏幕内容　出处：《中国图象图形学报》2017年07期

【摘要】：目的调色板编码是屏幕内容编码的典型方法之一,其索引图的编码效率直接影响到调色板编码算法的总体压缩性能。但是,在处理物体前景和文字边缘的过渡区或连接区索引时,现有索引图预测编码方法的效率仍有待改善。为此提出一种基于马尔可夫模型的索引图预测算法。方法随机选取了2 000个局部预测失败的索引值并将它们划分为3类典型分布,发现前2类分布的索引值往往处于边缘的灰度平滑过渡区,相邻索引值间呈现较为明显的线性变化,进而提出采用1阶2维马尔可夫随机过程来刻画这种线性性。对于一个待预测索引值,首先利用1阶2维马尔可夫模型计算相邻索引值的线性相关得到初始预测值,再利用颜色转移概率最大化确定其最优预测值。结果本文算法的预测准确率为97.53%,比多级预测算法(MSP)和基于局部方向相关性的预测算法分别平均提高了4.33%和2.10%,尤其适用于包含大量文字字符和几何图元的视频序列的索引图预测。并且,渐近时间复杂度与基于局部方向相关性的预测算法相当,明显低于MSP。具体地,本文算法的实际运行时间比MSP算法节省了95.08%,比基于局部方向相关性的预测算法增加了35.46%。结论本文提出的基于马尔可夫模型的索引图预测算法通过发掘索引值在边缘区域的线性相关性和特定的颜色转移模式,提高了索引预测效率,并保持了较低的计算复杂度,可应用在屏幕内容文本/图形块的调色板编码中。
[Abstract]:Objective the palette coding is one of the typical methods of screen content coding. The coding efficiency of the index graph directly affects the overall compression performance of the palette coding algorithm.However, the efficiency of the existing index prediction and coding methods is still improved when dealing with the index of the transition region or the link area of the foreground of the object and the edge of the text.Therefore, an index graph prediction algorithm based on Markov model is proposed.Methods two thousand index values of local prediction failure were randomly selected and divided into three types of typical distributions. It was found that the index values of the first two types of distributions were usually in the gray level smooth transition region of the edge.The linear variation between the adjacent index values is obvious, and the first-order 2-dimensional Markov stochastic process is proposed to characterize the linearity.For an index value to be predicted, the linear correlation of the adjacent index value is calculated by using the 1-order 2-D Markov model, and the optimal prediction value is determined by maximizing the color transfer probability.Results the prediction accuracy of this algorithm is 97.53, which is 4.33% and 2.10% higher than that of MSP-based multilevel prediction algorithm and local direction correlation algorithm, respectively. It is especially suitable for the prediction of video sequences with a large number of text characters and geometric elements.Moreover, the asymptotic time complexity is similar to the prediction algorithm based on local direction correlation, which is obviously lower than that of MSPs.Specifically, the actual running time of this algorithm is 95.08 less than that of MSP algorithm, and 35.46% more than the prediction algorithm based on local direction correlation.Conclusion the proposed index graph prediction algorithm based on Markov model can improve the efficiency of index prediction by exploring the linear correlation of index values in the edge region and the specific color transfer mode, and maintains a low computational complexity.Can be used in screen content text / graphics block palette coding.
【作者单位】： 辽宁师范大学计算机与信息技术学院;大连理工大学计算机科学与技术学院;
【基金】：国家自然科学基金项目(61402214,41271422) 高等学校博士学科点专项科研基金项目(20132136110002) 辽宁省教育厅科学研究一般基金项目(L2015285,L201683681) 大连市青年科技之星项目支持计划基金项目(2015R069)~~
【分类号】：O211.62;TN919.81
，

本文编号：1709576