Research On The Parallel Key Technology Of Image Feature Extraction Based On GPU

With the development of information technology,higher requirements for the performance and accuracy of image recognition are put forward.How to improve its performance is one of the hot research fields.The parallel acceleration scheme by GPU has become the focus of this field.In this paper,the relevant steps in the process of image recognition are studied,including the process of image preprocessing and the feature extraction and classification process of the image.This paper takes the face image recognition as the research object,GPU parallel computing was used to accelerate the key steps in the process of image recognition.First of all,in the preprocessing stage of image,we accelerate the process of image preprocessing by parallel optimization of filtering operation and down sampling operation in correlation image preprocessing algorithm.The pre processing methods effectively improve the recognition effect of the image feature extraction and classification stage.Secondly,in the stage of image feature extraction,we studied the convolutional neural network model based on multi GPU.We divided the convolutional neural network and deployed multiple volume layer networks on each GPU to carry out cyclic alternating training from the two aspects of data parallelism and model parallelism.The hybrid parallel alternation model effectively improves the utilization of computing resources,reduces data transmission between modules,and improves the training efficiency of convolutional neural networks.The training model of convolutional neural network based on multi GPU and the image processing based on GPU used in this paper develops the advantages of GPU in general calculation compared with the traditional image feature extraction scheme.In a large number of cases,it has high efficiency and good expansibility.The optimization of the whole image feature extraction process is effectively implemented.