Image Feature Extraction Methods

ATR is one of the most significant requests, although it is also one of the most challenging tasks. During past several decades great progress has been made in research on this subject. However, it is far away from satisfactory requirements from real world. ATR involves many techniques, such as Image preprocessing; Image enhancing; Image Segmentation; Feature extraction; classifiers designing and so on. Feature extraction is crucial. On one hand, researchers attempt to work out algorithms and methods to some special targets with high right classification rate and good efficiency. Among them, Principal Component Analysis, Fisher's Linear Discriminant, nonlinear algorithms mainly appearing as Kernel approaches, and so on. On the other hand, in real application efficiency is also an important indicator to assess one algorithm, because in many cases only algorithms with high efficiency can satisfy request of real task. This paper aims at designing feature extraction algorithms on face recognition, including linear feature extraction and nonlinear ones.Feature extraction approaches are divided into two groups in this paper, linear feature extraction and nonlinear feature extraction. The information after linear mapping is called linear features; the information after nonlinear mapping is called nonlinear features. The mappings are called linear feature extraction and nonlinear feature extraction correspondingly.Principal Component Analysis and Fisher's Linear Discriminant are two methods widely used. This paper introduces feature extraction approaches, 2DPCA and 2DFLD, respectively. We develops the 2DFLD, and presents a new feature extraction approach called blocked FLD. 2DFLD is the special case of blocked FLD. the experimental results indicated that the recognition performance of blocked FLD is superior to that of 2DFLD.Kernel method is a powerful machine learning method developed recently. It builds on the statistical learning theory. Feature extraction based on kernel is discussed in detail. A feature fusion method combined with KPLS is proposed.