Canny算子在DSP上的实现及优化

摘要

In order to realise the quick image segmentation in the DSP platform,the Canny detector theory is analyzed in detail. The arithmetic transplant was implemented in combition with TI DSP TMS320C6678 processor properties. Aiming at data interaction with the images in external memory,the past mode that the images were accessed according to gray values one by one was improved. The vectorized packed data was designed for dealing with Gaussian filtering to improve the parallel operation, and the accessing method of wide memory in the process of gradient compution and the threshold value calculation was adopted to improve the read-write efficiency of external memory. The result indicates that the optimization method designed in this paper can improve the velocity of Canny operator under the condition of not changing the segmentation result,and can provide a reference for the engineers to carry out arithmetic transplant and optimization on the DSP platform.%为了在DSP平台实现细胞图像快速分割，详细分析Canny算子原理，结合TI DSP TMS320C6678处理器特性，实现了算法移植。针对与外部存储器图像数据交互，改变以往对图像逐灰度值进行访问的方式，设计了矢量化数据打包方法处理高斯滤波来提高并行运算。且在梯度计算、阈值计算过程中，采用宽存储器访问方法提高读写外部存储器效率。结果表明设计的优化方法在不改变分割效果前提下改善了算子速度，可为工程人员在DSP平台进行算法移植与优化提供借鉴。