基于FPGA的快速图像纹理特征提取方法的研究

发布时间：2018-08-11 17:16

【摘要】：图像特征的一个重要分支就是纹理特征,它体现了不同图像和物体的形态、大小、分布、方向等重要参数,对图像特征识别起到决定性因素。但是纹理特征提取的过程十分复杂且计算量巨大,为了解决这个难题,提出了一种在现场可编程逻辑门阵列(FPGA)平台下实现纹理特征提取新方法。首先对基本图像特征算法做了并行化的优化,从算法的数值范围和表示精度两个角度,做了相应的分析和误差控制,从而适应FPGA的运行。然后对FPGA的数据流传输提出了一个高效率的解决办法,该方法对其中的主要模块采用了流水线优化,并采用寄存器配置模式,从而在线地修改参数,适应不同的图像大小和卷积核等环境变量。结果表明,在同等功耗条件下,可以达到10倍于CPU的性能,达到了快速提取特征的目的。
[Abstract]:An important branch of image feature is texture feature, which embodies important parameters such as shape, size, distribution and direction of different images and objects, and plays a decisive role in image feature recognition. But the process of texture feature extraction is very complicated and the computation is huge. In order to solve this problem, a new method of texture feature extraction based on field programmable gate array (FPGA) platform is proposed. Firstly, the basic image feature algorithm is optimized by parallelization, and the corresponding analysis and error control are made from the point of view of the numerical range and the representation accuracy of the algorithm, so as to adapt to the operation of FPGA. Then, a high efficient solution to the data stream transmission of FPGA is proposed. The main modules are optimized by pipeline and register configuration mode, which can modify the parameters online. It adapts to different environmental variables such as image size and convolution kernel. The results show that the performance of CPU is 10 times higher than that of CPU under the same power consumption, and the purpose of feature extraction is achieved quickly.
【作者单位】：南京信息工程大学江苏省气象探测与信息处理重点实验室;南京信息工程大学江苏省大气环境与装备技术协同创新中心;南京信息工程大学电子与信息工程学院;
【基金】：国家自然科学基金(11202106) 江苏省气象探测与信息处理重点实验室开放课题(KDXS1401) 江苏高校品牌专业建设工程江苏省“信息与通信工程”优势学科建设项目资助
【分类号】：TP391.41

【相似文献】