Performance Analysis of Morphological Operations in Cpu and Gpu for Accelerating Digital Image Applications

In this paper, we evaluate the performance of morphological operations in central processing unit (CPU) and graphics processing unit (GPU) on various sizes of image and structuring element. The languages selected for algorithm implementation are C++, Matlab for CPU and CUDA for GPU. The parallel programming approach using threads for image analysis is done on basic entities of images. The morphological operations namely dilation and erosion are purely depends upon local neighborhood information of each pixel and thus independent. GPU capable to create more number of threads. Here thread per pixel of the image is created to execute the algorithms with the neighbors of relative pixels. Finally the speed performance of all algorithms on conventional processor CPU and parallel processor GPU are computed and compared. Dilation operation in GPU is up to 5 times faster than CPU C++ code and up to 4 11 times faster than CPU MATLAB code, likewise Erosion operation in GPU is up to 2 times faster than CPU C++ code and up to 612 times faster than CPU MATLAB code when image size varying from 256 × 256 to 1024 × 1024. Further it shows that the performance of GPU implementation is gearing up when the image size is increased. While changing the structuring element size with 1024 × 1024 image the Dilation operation in GPU is up to 3-5 times faster than CPU C++ code and up to 10 35 times faster than CPU MATLAB code, likewise Erosion operation in GPU is up to 2 – 6 times faster than CPU C++ code and up to 1246 times faster than CPU MATLAB code.