Algorithm Research Of 2DNMF Based On Conjugate Gradient

Face recognition is based on the analysis of facial features, compare feature information to identify the identity. The method of Face recognition based on subspace analysis, which is concerned by researchers in recent years to extract face feature one of the methods. But in the process of feature extraction and recognition judgment, negative value is not easy to explain, even no physical significance. So people turn to consider the method which based on subspace analysis of nonnegative matrix decomposition algorithm (NMF), through introducing a nonnegative constraint, to extract the face local, sparse characteristics, not only reflects the local form whole thought, and easy implementation, decomposition results not negative, interpretability is strong, with clear physical meaning. However, the traditional NMF algorithm uses the rules of the multiplicative iterative alternating update. This iteration is essentially to design based on the gradient descent method, the convergence speed is slow, for the training of the feature extraction for a long time, it is difficult to adapt to the mass data processing.In the comprehensive research of various kinds of two-dimensional nonnegative matrix decomposition algorithm, on the basis of taking into account the human face image recognition characteristics under the condition of shade, noise, this paper puts forward two improved algorithm based on 2DNMF:(1) In view of the traditional 2DNMF multiplicative iterative formula used in the decomposition, using conjugate gradient method based on matrix form combined with the alternate iteration to get the new iterative format; Again to the two base to category index in the form of a kind of weighted sum for face recognition.(2) In the objective function by matrix orthogonal constraints, deduces a new iterative format, to replace the traditional multiplicative iterative; Again to the two categories on the basis of weighted sum of the index, for face recognition.Improved algorithm fully considered the facial image information, extract information to achieve the maximum fusion respectively, so as to improve the recognition rate. Experiments show that the proposed improved algorithm in covered the face image recognition obtained the quite ideal effect.