Research On 3D Face Reconstruction And Recognition Based On Photometric Stereo Technology

Three-dimensional reconstruction technology has always been a major constraint to the development of 3D face recognition technology.Among many 3D reconstruction technologies,photometric stereo technology is popular among people because of its high reconstruction accuracy,low reconstruction cost and simple operation.The traditional photometric stereo technology uses parallel light,and the application range of the traditional photometric stereo technology is very small due to the limitation of parallel light.Therefore,in order to enable the photometric stereoscopic technology to be applied to a wider range of fields,the subject uses a point source and a line source instead of parallel light for photometric stereo technology.This topic deeply analyzes the illuminating model of the point source,and finds that the general led point source is an anisotropic source,and its illuminance is not only attenuated with distance,but also attenuated with the illuminating angle.The experimental results show that the accuracy of the reconstruction result of the point source according to its true illumination model is significantly higher than that of the point source as the ideal point source.In the reconstruction process,the direction of the light source needs to be calibrated.The traditional calibration method needs to place the reference object.This paper proposes a laser calibration method to realize the automatic calibration of the point source direction.In order to improve the accuracy of the reconstruction results,this paper proposes an iterative algorithm and a shadow processing algorithm for 3D reconstruction.For a hemispherical surface with a diameter of 40 mm,the highest reconstruction accuracy of the point source luminosity stereoscopic technique proposed by the subject is 0.63 mm.When the point source is at different positions,the image noise has different effects on the reconstruction result.Therefore,the subject also analyzes the constraint conditions that the optimal position of the point source should satisfy.When the point source is in the optimal position,the image noise has the least influence on the reconstruction result.The intensity of a conventional line source illuminating a certain point needs to be calculated by integration.When the image resolution is relatively high,the amount of calculation required for the luminosity stereo technique based on the conventional line source is very large.Therefore,in order to improve the speed of reconstruction,this paper proposes a stripline light source composed of 8 led point light sources.This line source can be approximated as a point source,which greatly simplifies the calculation process of the line source photometric stereo technology.For a hemispherical surface with a diameter of 40 mm,the highest reconstruction accuracy of the line source photometric stereoscopic technique proposed by this subject is 1.13 mm.Finally,this topic establishes a 3D face database using point source photometric stereo technology,and then uses the database to train the improved AlexNet network.The improved AlexNet network has a recognition accuracy of 92% after 10,000 training sessions.