基于交叉熵的倾斜文本图像细节特征提取仿真  

Simulation of Feature Extraction of Tilted Text Image Based on Cross Entropy

作　　者：张润 冯云霞[1] ZHANG Run;FENG Yun-xia(School of Information Science and Technology,Qingdao University of Science and Technology,Qingdao Shandong 266100,China)

机构地区：[1]青岛科技大学信息科学技术学院,山东青岛266100

出　　处：《计算机仿真》2020年第7期489-492,共4页Computer Simulation

摘　　要：针对传统的倾斜文本图像细节特征提取方法存在着特征提取时间长,特征辨识力低等问题,提出一种基于交叉熵的倾斜文本图像细节特征提取方法,通过交叉熵方法来构建倾斜文本图像灰度共生矩阵,运用并行算法与并行体系结构对灰度共生矩阵进行计算,获取矩阵的统计特征,利用矩阵中元素之间计算的并行性,对各个像素进行并行计算。根据计算结果来设定倾斜文本图像纹理二阶矩、熵、对比度以及均匀性等多种纹理特征向量,分析其纹理特征向量,结合分析结果对倾斜文本图像灰度共生矩阵的特征进行提取。实验结果证明,所提方法可以有效减少特征提取所需要的时间,提高倾斜文本图像细节特征辨识力。Traditionally,the feature extraction is time consuming and the feature recognition ability is low.Therefore,this article presented a method to extract the detail feature of skew text image based on cross entropy.First of all,the cross entropy method was used to construct the grayscale co-occurrence matrix of oblique text image.Then,the parallel algorithm and the parallel architecture were used to calculate the gray level co-occurrence matrix,so as to obtain the statistical feature of matrix.In addition,the concurrency among the elements in matrix was used to perform parallel calculation on each pixel.According to the calculation result,many kinds of texture feature vectors such as second-order moment,entropy,contrast and uniformity of oblique text image texture were given.Finally,the texture feature vector was analyzed and the features of grayscale co-occurrence matrix of oblique text image were extracted.Simulation results show that the proposed method can effectively reduce the time required by feature extraction and improve the feature recognizing ability of skew text image detail.

关 键 词：灰度共生矩阵 并行算法 纹理特征 并行体系结构 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]