共生矩阵纹理特征的DSP优化实现

DSP Optimized Implementation of Co-occurrence Matrix Texture Feature

灰度共生矩阵纹理特征具有计算复杂、耗费时间等问题,严重影响了程序执行的效率。针对此问题,分析了共生矩阵纹理特征的原理,研究了TMS320C6678 DSP的结构性能,提出了基于共生矩阵纹理特征的存取带宽和软件流水的优化方法,在CCS5.3软件平台下选择TMS320C6678对其进行了程序实现,最后使程序执行时间从1.94 ms减少到0.259ms。实验结果表明,提出的优化方法能够缩减代码执行时间,提升代码性能,满足嵌入式图像处理系统的实际需要。

Texture features based on Gray Level Co-occurrence Matrix （ GLCM） has the drawbacks of complex calculations and time-consuming,which seriously affects the efficiency of program execution. Aiming at this problem,analyze the principle of the extracted tex-ture features based on GLCM in this paper,study on the performance of the TMS320C6678 DSP,propose the optimization method of ac-cess bandwidth and software pipelining based on GLCM texture feature,then achieve the optimization of GLCM-based texture features by using of TMS320C6678 in the CCS5. 3 software platform,so that the program execution time is reduced from 1. 94 ms to 0. 259 ms. Ex-periment results show that the proposed optimization method can shrink code execution time, improve code performance and meet the needs of embedded image processing system.