新的图像几何矩快速算法

Novel Fast Computation of Image Geometric Moments

矩技术作为一种有效的图像描述方法,在图像分析方面有着重要应用,但由于涉及大量计算,在一定程度上制约其应用.提出一种新的基于GPU的快速计算方法,它首先将图像加载成纹理,然后在像素着色器中利用GPU内核同时对多个像素进行处理,快速计算所需矩值.实验结果表明,与当前的直接法和对称核快速算法相比,文中提出的算法可有效节省计算时间,尤其在图像尺寸较大及所使用的矩的阶数较高的情况下,计算性能更优.

As an effective image description method,moment techniques have been used in image analysis.However,their computation limits their application when real-time contains are imposed.This paper proposes a novel algorithm for high performance computation of image geometric moments based on GPU（graphic processing unit） by making use of its parallel computation capability,programmable and general purpose computation.After original input image loading texture,the steam programming computes geometric moments in pixel shader.Experimental results show that the proposed method outperforms the existing software implementation approaches,such as direct method,symmetric kernel method based on CPU and so on,especially for larger images and higher order moments which can be performed in real-time.