Research And Implementation Of Face Image Retrieval Based On PCA And LDA

With the development of information technology, in scientific research, industrial production, agricultural planting and other fields, the visualization data such as image and video are growing rapidly. At the same time, compared with the traditional information retrieval methods, image retrieval and video retrieval become more important. Face is one of the important biological characteristics of individuals and important visual objects of image. The face retrieval technology is widely used, and becomes a research hotspot in the field of Information Science in recent years. In this paper, some research is done about the face image retrieval technology in the following aspects.(1) The similarities and differences between the face retrieval model and the face recognition model are researched and analyzed. And the traditional face retrieval model is improved, which is suitable for face recognition algorithm based on statistical characteristics and improves the practicability of the system.(2) In face recognition process, when extract test and training face feature and calculate the distance between features, we need a method to determine the test images in the category. The nearest neighbor classifier and K nearest neighbor classifier is two kinds of commonly used classification methods, but the change of illumination, expression and other external factors will lead to these two kinds of classifier misjudge. Based on nearest neighbor classifier and K nearest neighbor classifier, a new classifier called custom classifier is proposed. The classifier selects several candidate images which have the nearest distance and compared there’s category label with the test image category label, the results for the image really image category as the recognition results. Experiment shows that the classifier can significantly improve the correct rate of recognition algorithm.(3) The analyzed methods based on Fisher discriminate are studied. Based on Fisher discriminate and 2DPCA algorithm, a new face recognition method is proposed. The new algorithm firstly uses the 2DPCA method to extract face feature, and then use the LDA criterion to select feature. Compared with traditional 2DPCA and LDA algorithm, the algorithm with the custom classifier has higher recognition rate.(4) According to the improved model of the face retrieval system, using the face detection algorithm based on color and the new face recognition method based on Fisher and 2DPCA, a face image retrieval system is designed and implemented.