期刊名称：Lecture Notes in Engineering and Computer Science
印刷版ISSN：2078-0958
电子版ISSN：2078-0966
出版年度：2018
卷号：2231&2232
页码：459-464
出版社：Newswood and International Association of Engineers
摘要：The traditional haar-like feature extraction
algorithm is a method based on integral image to help
extract image features with different feature modules.
However, there are many problems with this kind of
method: there are too many features extracted, there is
redundant information and the expression of the target
information is not enough. In view of these shortcomings,
the paper adopts the improved active appearance model
(AAM) to extract the image features, and compresses the
multidimensional feature information by using the
compression sampling method. The recognition
classification uses the Adaboost classifier training method
to the compressed feature space. Experiments show that the
training time required by the classifier is reduced by
compressing the extracted eigenvalues, and the recognition
performance is also better than the traditional algorithm.
