Face identification method based on independent component analysis network

Abstract

The invention relates to a face identification method based on an independent component analysis network. The method comprises the following steps: 1) carrying out cutting, aligning and normalizationpretreatment on input face images; 2) carrying out filtering through a group of trained ICA filters to obtain a group of mapping images; 3) carrying out non-linear and pooling processing on each mapping image to obtain a more efficient feature mapping image; 4) carrying out block LBP coding on each mapping image, and then, stringing the obtained local features to obtain feature expression; and 5)carrying out WPCA dimension reduction on the feature expression, and finally, carrying out identification verification on the two pieces face images through a cosine similarity measurement method. Thetrained ICA filters are applied to a CNN to form a single network; multi-scale information can be obtained based on different sensing areas of the ICA filters; and the method can ensure higher recognition rate in face recognition, and meanwhile, reduces calculation amount effectively and is convenient for popularization and application.

Description

technical field [0001] The invention belongs to the field of computer vision, in particular to a face recognition method based on an independent component analysis network. Background technique [0002] Face recognition is a very important part in the field of computer vision, and plays an important role in information security, video surveillance, etc. The face recognition system mainly includes three parts: (1) face detection; (2) face feature expression; (3) face recognition (feature classification). Among them, good face feature expression plays a very important role in face recognition. [0003] Face feature expression mainly goes through three periods: (1) global feature expression: the main representative methods are Eigenfaces and Fisherfaces methods, the aforementioned methods all use the features of the whole face as input; (2) local feature expression: the main methods are LBP (Local Binary Patterns), SIFT (Scale Invariant Feature Transform), Gabor, HOG (Histogr...

AI Technical Summary

Problems solved by technology

However, CNN, as one of the supervised learning methods, needs to learn from a large number of training samples of calibration type information. Although there are many public face databases, the training takes a lot of time and requires higher hardware configuration. At the same time, the setting of hyperparameters manual experience

Images