Abstract
A principal components analysis (PCA) algorithm is one of the most important algorithms that has been used for doing many tasks; for example, data dimension reduction, data compression such as image compression, pattern recognition such as face detection and recognition, and many other things. An improved principal components analysis (IPCA) algorithm is similar to the PCA algorithm except that it uses the concepts of Shannon information theory for improving the PCA algorithm. It has been claimed that the IPCA algorithm behaves better than the PCA algorithm. Due to the huge importance of the PCA algorithm where it is commonly used, we were motivated to theoretically and empirically compare the behavior of the PCA and IPCA algorithms in different applications.
This paper validates the IPCA algorithm on images for the first time where it has not been tested on images before. It also proposes a new learning method for face detection and recognition using the PCA and IPCA algorithms. In addition, this paper evaluates the performance of the PCA algorithm versus IPCA algorithm in image compression, and in face detection and recognition where we have obtained rigorous decision about which algorithm behaves better in each area. We have also proposed a new method for evaluating the performance of the PCA and IPCA algorithms in image compression based on three measures. We finally have proposed to use another segmentation method with the algorithms in order to center and normalize only pixels that occupy faces for obtaining better performance.
Lastly, the MATLAB software has been used for performing our experiments. We have found that the PCA algorithm, in general, behaves better than the IPCA algorithm in the most of the areas. It is better than the IPCA algorithm in face detection and recognition. The PCA algorithm is slightly slower than the IPCA algorithm, but it has significant small error rates. Also, it is easier in computation than the IPCA algorithm. On the other hand, the IPCA algorithm is better than the PCA algorithm in image compression because it obtains higher compression, more accurate reconstruction, and faster processing speed with acceptable errors.