Image Segmentation Based On Membership In The Value Of PCA And LDA Face Recognition Improvements

With the rapid development of technology, many areas have been pressing needs fo r face recognition technology, and many ways of the recognition have become hot resear ch focuses of scholars, including subspace analysis method which has received more att ention for its good effect and less calculation. In this paper, this method is also conducte d in-depth studies. The research work has been done as follows:Principal component analysis (PCA) is the best two-dimensional image compressio n optional orthogonal transformation. It first changes the matrix which makes up image into a vector, then calculates the covariance matrix for feature extraction. The number of dimensions is generally high, so it will result in a problem of small sample. This paper p roposes FGPCA aiming at the problem, which first proposes characteristics of each part of the image and calculates the local features of each training sample for membership, t o achieve partial images classification, and draw a complete picture of the test to set the degree of membership for their training through the membership of the weighted sum. T hus it improves the membership function, making the recognition rate higher.The traditional method of LDA exaggerated the contributions to the triage of distan t classes and sample, which existed sub optional problem. You can not identify the most favorable classified projection vector to affect the classification results of LDA. Therefo re, KDA algorithm in the text abandons the whole class mean in between class scatter m atrix using the mean of parts of the class so that the center of the class will not deviate; within-class scatter matrix gives a weighting to each training sample in order to distingu ish the primary and secondary contribution between the different samples to avoid dista nt sample dominating; MDA algorithm further improves the calculation of the mid-valu e, so that it not only retains more image information but also reduces the impact of outli ers. Finally, integration of the two methods leads to MKDA algorithm. This algorithm is a good solution to the sub optional problem and it enhances robustness.