Small sample face recognition method based on sliding block generative adversarial network

Abstract

The invention discloses a small sample face recognition method based on a sliding block generative adversarial network. The small sample face recognition method is used for emphatically processing posture changes. The method comprises the following steps: firstly, performing mirror image transformation and symmetric transformation on a front face-side face sample in an original data set to generate a virtual new sample; subjecting all the samples to cross alignment treatment to guarantee the accuracy of sliding partitioning; changing the overall features of the human face into overlapped localfeatures by adopting a sliding window, then sequentially and orderly inputting the overlapped local features into a generative adversarial network, and generating a front human face by utilizing an improved combined loss function; and finally, establishing a generation and recognition integrated system to realize small sample face recognition under posture change. According to the invention, therequirement of the network for the number of training samples is greatly reduced, the method has certain effectiveness for face images with illumination changes and small posture changes (the angle issmaller than 45 degrees), and a competitive recognition effect is achieved on the small sample face problem under posture changes.

Description

technical field [0001] The invention belongs to the technical field of face recognition, and in particular relates to a small-sample face recognition method based on a sliding block generation confrontation network. Background technique [0002] In recent years, the demand for biometric technology in life, finance, law, criminal investigation and other application scenarios has been increasing, and face recognition technology has become the most popular biometric technology due to its rich feature information, easy collection and high precision. technology. Face recognition systems in practical application scenarios are mostly small-sample problems, that is, only a few samples can be collected for each face category in the face database. We collectively refer to this type of problems as small-sample face recognition problems (FFR). Usually, the collection of face images is based on uncontrollable natural conditions, often with various changes in illumination, expression, po...

AI Technical Summary

Problems solved by technology

Usually, the collection of face images is based on uncontrollable natural conditions, often with various changes in illumination, expression, posture, occlusion, etc. Usually, the collection environment of face images is in an uncontrollable natural environment Face samples often contain changes such as illumination, posture, occlusion, expression, noise, etc., and these changes will affect the recognition accuracy of the system to a certain extent.

Images