基于CUDA的红外图像快速增强算法研究

针对红外图像边缘模糊,对比度低的问题,文中研究了改进的中值滤波和改进的Sobel边缘检测对红外图像进行处理。在对处理后图像的特征进行分析的基础上,研究了改进的Laplace金字塔分解的图像融合算法,并基于CUDA并行处理技术,在可编程GPU上实现了红外图像快速增强的目的。该算法结合GPU的内存特点,应用纹理映射、多点访问、并行触发技术,优化数据的存储结构,提高数据处理速度,适用于对红外图像增强的实时性要求较高的领域。实验结果表明,该算法有较好的并行特性,能充分利用CUDA的并行计算能力,提高了红外图像增强的实时性,处理分辨率为3 096×3 096的红外图像时加速比达32.189。

Research of fast infrared image enhanced algorithm based on CUDA

To deal with the problem of edge blur and low contrast,of the infrared image in infrared image processing,improved median filtering and improved Sobel edge detection are discussed.After the image being processed by algorithms of improved median filtering and improved Sobel edge detection,the features of the image are analyzed so that the improved image fusion algorithm of Laplace pyramid is presented.Programmable GPU is used to realize the purpose of fast infrared image enhancement based on CUDA parallel processing technology.The improved image fusion algorithm of Laplace pyramid is suitable for high real-time require areas of infrared image enhancement because that texture mapping,multi-access and parallel triggering technology are combined with characteristics of the GPU memory to optimize the data storage structure and speed up the data processing.Experimental results show that the algorithm has good parallel properties and can improve the real-time of infrared image enhancement through effectively implementing CUDA parallel computing power.The speedup is up to 32.189 in the 3096 × 3096 infrared image processing.