Featuring extraction using independent component analysis

Abstract

For several decades, many researchers have studied for extracting features from data and classifying the patterns using them. Above all, the research about independent component analysis (ICA) is noticeable because we can get information from data by imposing the nature of independence on them. The goal of our work is to compress the data into simple structures and then express them as exact as possible. Herein, we use two methodology for compressing the data. Firstly, we use principal component analysis (PCA). This method compress the data by using the eigenvectors of input correlation matrix. The Second is kirsch edge detection which detects the directions of data components and if we use this with PCA, we can considerably reduce the dimension of data. We focused on determining the principles of classification by extracting the features of independent components. To test the proposed method, we experiment the performance of handwritten digits recognition (HDR) using USPS database, which has total 10 classes from 0 to 9. In this study, we applied new frameworks using ICA for efficient data recognition and evaluated our approach through HDR experiments. From the experimental results, we have shown that the proposed method can generate effective features for pattern recognition. And the suggested feature extraction techniques can be applied to compression, reconstruction, code-making, and recognition of the data.