Comparative study of linear and nonlinear color model identification based optimal feature extraction

Feature extraction involves detecting and separating fundamental portions of digital images, as well as simplifying a complex input data set into a set of features in a reduced order. It has a variety of practical applications in virtually all areas of pattern recognition and image processing disciplines. Dimensionality reduction is a major goal of feature extraction to retain the relevant information and eliminate the redundant and noisy information. To accurately describe a huge set of complex data with a lower dimensional space representation, optimality should be introduced into feature extraction, which includes both linear and nonlinear approaches, among which, both Principal Components Analysis (PCA) and Independent Component Analysis (ICA) are the most successful ones. Merits and drawbacks of these two linear and nonlinear approaches can be obtained via comparative analysis. The RGB model is an additive color model which displays a color by mixing intensity levels of red, green and blue components. The grayscale value is set when all three primary colors have the same intensity. Most digital images are displayed in RGB format, so optimal feature extraction should be conducted according to three individual components in the RGB model. In order to evaluate the two leading schemes of PCA and ICA for feature extraction, the quantitative metrics based on the information theory have been introduced so as to examine the qualities of optimal feature extraction approaches from both subjective and objective points of view.