The Survey Of Dimensionality Reduction In The Face Recognition Algorithm

Now many feature extraction and face recognition algorithms, subspace analysis algorithm for its simple calculation, separability characteristics such as good and effective by people's extensive concern. The basic idea is based on certain performance goals to find a linear or non-linear space transformation, the original data compression to a low-dimensional subspace. This paper mainly around subspace method of feature extraction and its application in face recognition from the traditional subspace unfolds, and zhang quantum space algorithm, this paper mainly studies the following research work:1. As often in high dimensional data fields related scientific and industry, such as computer vision and pattern recognition, biological information and aerospace, etc. When we deal with these figures, they will often be of high dimensional attribute processing and using these data, which is reflected in the obstacles and their computational complexity is higher and the result is not optimal. Dimension reduction is reduced by high dimensional data about the low dimensional process and used to reveal the nature of low dimensional structure data. It as to overcome "dimension disaster" approach in the related field plays an important role. In the last few decades, have a lot of dimension-reduction methods were proposed unceasingly and in-depth research. Through summing up and compare the currently popular linear dimension reduction method are given, and the future development direction of traditional dimension reduction.2. Nuclear method is newly developed a kind of pattern recognition method, its theory based on statistical learning theory. Statistical pattern recognition in solving plays a basic role, but the traditional statistics achievements of mostly based on the asymptotic theory, namely sample observation number above when an infinite number of tends to statistical properties. And based on kernel methods data dimension reduction is currently processing multi-dimensional data, but also an important step in a heavy machine learning research topic. Detail several typical nonlinear reduced-order method and algorithm of time complexity from two aspects of advantages and disadvantages of these algorithms of the in-depth analysis and comparison. Finally proposed nonlinear data dimension reduction to the problem still.3. The image actually exist in tensor, which itself contains information not vector can be completely replaced. When Vasilescu and Terzopoulos face recognition algorithm proposed tensor Tensorfaces, really will face recognition image from traditional vector algorithm to tensor, singular value decomposition of the tensor space expands, the high order tensor said according to different dimension of face image for light, the direction of decomposition factors such as expressions, gestures, facial recognition more accurate and convenient. For existing tensor type learning algorithm for vector model summarizes: said existing problems, discusses design tensor in existing in face recognition algorithm and the necessity and the meaning of existence.