Image Feature Extraction Technology Based On Incremental Two-Dimensional Principal Component Analysis

It is important for image processing technology to extract important and valuable essential features from the data,and remove irrelevant and redundant information.Principal component analysis(PCA)can describe the high-dimensional data by selecting a few principal components which can highlight the similarities and differences between the data to reduce the dimension of the data.According to the way of obtaining the data,the principal component analysis is divided into a batch principal component analysis and an incremental principal component analysis.The batch principal component analysis focus on the sample data's covariance matrix,therefore,they require all the image data before analysis.The incremental principal component analysis does not need to obtain all image data at once,it uses incremental learning method to iteratively update the estimated value of the principal component,avoiding direct calculation of the covariance matrix.With the rapid improvement of the camera equipment resolution and the continuous progress of data storage technology,it will lead to the large covariance matrix and high computational cost of traditional batch principal component analysis,which makes the feature extraction speed not keep up with the data update speed,and the incremental principal component analysis method is more suitable.The main contents of this paper are summarized as following:(1)Two-dimensional principal component analysis algorithm(2DPCA)can be performed in the batch mode and can not meet the requirements of feature extraction for high-dimensional mass image data and on-line data.To overcome these limitations,the incremental learning of the candid covariance-free incremental PCA(CCIPCA)is innovated to the existing 2DPCA,and the called incremental 2DPCA(I2DPCA)is firstly presented to incrementally compute the principal components of a sequence of samples directly on the 2D image matrices without estimating the covariance matrices.Therefore,the I2 DPCA can improve the feature extraction speed and reduce the required memory.(2)The proposed I2 DPCA algorithm only extracts the row direction feature of the image matrix,however,the variations between the column direction,generally neglected,are also useful for the high-accuracy object recognition.It will lead the high extracted feature dimension,and the dimensionality reduction effect is not satisfactory.Thus,another incremental sequential row-column 2DPCA algorithm(IRC2DPCA)is proposed.The IRC2 DPCA can compress the image matrices in the row and column direction,and the feature matrices extracted by the IRC2 DPCA are with less dimensions than the I2 DPCA.It can effectively reduce the time and memory in the classification process.In this paper,we selected the block database,the ORL database and the Yale database as the experimental samples,and carried out the experiments of the principal component convergence,the classification rate,the computation complexity and the image reconstruction respectively.The substantial experimental results show that the two incremental algorithms I2 DPCA and IRC2 DPCA proposed in the paper are effective.In addition,they can surpass other PCA algorithms with the performance of the computation time and the required memory.