Research and implementation of face recognition system based on cloud computing

Research and implementation of face recognition system based on cloud computing

South Lee Shi

The main work done in this paper is as follows: 1. With 5 machines in master/slaver structure, the network and files are configured in Linux system, and the relevant software is installed, so as to complete the construction of the Hadoop distributed system. 2. Based on the whole structure of face recognition, this paper analyzes the different common face recognition algorithms, such as feature extraction and classifier design, which can adapt to the MapReduce computing framework under Hadoop. The criterion of whether the MapReduce is suitable for the calculation framework is whether the calculated degree of separation is high, and whether the calculation amount is small enough. 3. Apply hog features to face recognition. The experimental results show that the hog based on gradient histogram can adapt to various illumination. 4. The SVM algorithm is often used in two classifiers. And SVM itself has two disadvantages: 1. Only the kernel function can be used in the case of linear irreducible condition, and the choice of kernel function lacks the effective theory to support it. 2. It is difficult to classify the distance between the classification plane and the boundary sample with only one classification plane. In order to solve these two problems, this paper proposes the svm-adaboost algorithm and the NSMD algorithm, which combines the Non-nuclear SVM and the adaboost algorithm to solve the linear problem, the latter constructs the compound classifier by taking the classification plane in the middle of the nearest sample multiple times, This solves the two disadvantages of SVM at the same time. 5. The whole face recognition system is implemented on the Hadoop system, and the MapReduce framework is implemented for each algorithm in the system. Before feature extraction, it is necessary to use a mapreduce to preprocess each image. The feature extraction needs to undergo three mapreduce process, once used in hog, two times for PCA covariance matrix and dimensionality reduction mapping. On the classifier, it is only necessary to use a mapreduce for both recognition and training.

Research and implementation of face recognition system based on cloud computing