Study And Implementation Of CPU-GPU Parallel Multi-Image Homomorphic Filtering Algorithms

Image enhancement is important image preprocessing technology to study, how to enhance an image efficiently and quickly is one of the hot issues in recent years. Homomorphic filtering technology is known as a key algorithm of image enhancement which is based on the frequency domain. However, the algorithm contains The Fourier Transformation and a huge number of loops. Once the image resolution is too high, when a large amount of operations need to be done, the serial algorithm will be difficult to achieve the real-time enhancement, so the parallel algorithm should be used to improve the processing speed. Therefore, the studying and the designing of The Homomorphic filtering parallel algorithm for multi Core GPU/CPU systems has the important practical significance.This article designed and implemented the parallel Homomorphic filtering algorithm S-GPUTXLB for multiple images on the multi-core CPU/single-core GPU system based on proceeding Fourier transform and inverse Fourier transform for parallel Homomorphic filtering algorithms for multiple images by calling GPU kernel function library. Based on that, we used the independence of each image proceeding task, then we implemented the parallel Homomorphic filtering algorithms D-GPUTXLB for multiple Images on multi-core GPU Systems by allocating several images to several cores of GPU. The experimental results show that proceed the multiple images Homomorphic filtering parallel algorithm S-GPUTXLB and D-GPUTXLB by using the multi-core CPU/GPU machine can significantly shorten the running time of processing.This article proceeded with greedy strategy, and then find a way to task allocation for proceeding the homomorphic filtering parallel algorithm for multiple images on the Hybrid heterogeneous cluster, through calling and proceeding of the parallel homomorphic filtering algorithms for multiple Images on multi-core GPU/CPU systems at each node of the cluster.At last, we designed and Implemented the Parallel Homomorphic Filtering Algorithms D-GPUHCPA for Multiple Images on Multi-core CPU/GPU Systems on the Hybrid heterogeneous cluster.The experiment tested the algorithm D-GPUHCPA with different number of threads from the thread block in GPU. We processed the algorithm D-GPUHCPA on a hybrid architecture multi-core CPU/GPU Dawning 5000A cluster system, experimental result shows that the Parallel Homomorphic Filtering Algorithms D-GPUHCPA for Multiple Images on the Hybrid heterogeneous cluster described in this article is efficient.