Research On Image Recognition Based On ICA And ELM

In recent years, Independent Component Analysis algorithm (ICA) has been widely used in image research and application fields. It has many applications such as in the speech signal separation, image signal de-noising, face image recognition, financial data analysis, medical signal processing etc. Neural networks have been hardly used in real-time computational field for its time-consuming training and low efficiency. Since 2006, Extreme Learning Machine (ELM) has been proposed by Guangbin Huang as a useful tool to solve issues related to training time, which inspires a myriad of researchers to highlight neural networks again.This paper mainly deals with image feature extraction, image reconstruction, image compression, image recognition and so on. First, some current image recognition algorithms are introduced, and then it introduces ICA algorithm and ELM algorithm. This paper proposes a new algorithm combined ICA and ELM to extract image features for classification fast. In this paper, different databases and algorithms are used to compare their performance. At last, the high accuracy and efficiency of the new algorithm are certified. The main works are as follows:1) This paper describes ICA algorithm systemically, especially fast independent component analysis algorithm (FastICA). It uses ICA algorithm for feature extraction from natural image and face image, and complete image reconstruction and image compression goals, and then the ICA algorithm can be extended so that it can be applied to color images directly.2) This paper introduces ELM algorithm completely, and measures the performance of ELM algorithm using regression experiments on different data sets.3) This paper proposes a new algorithm combined ICA and ELM. The new method uses the traditional ICA algorithm to extract the common visual features from natural images, then utilizes ELM algorithm for real-time feature extraction of face image. Finally, ELM classifier is used for face image recognition. The experiments show that utilizing the fitting features by ELM algorithm has higher recognition rate and faster speed in feature extraction than directly using ICA algorithm. The data sets used in the experiments are Yale face database and MNIST handwritten digital database.Finally, a lot of Matlab simulation experiments are executed to validate the algorithm, and some enhancements of the algorithm are described in this paper. In addition, this paper discusses an expectation on the application of the new algorithm.