Research And Implementation Of A Multi-perspective Eyeglasses Virtual Try-on Algorithm Based On Feature Detection

With the development of Internet and e-commerce,shopping in the virtual network environment becomes more and more popular,but Internet shopping unlike the traditional shopping mode where people can try on personally,which often makes customers have cognitive bias over clothing,eyeglasses and other goods,resulting in a large number of return and exchange,causing a great waste of social resources.Since the virtual try-on technology can effectively reduce cognitive biases of commodity,the market has a huge demand for virtual try-on technology and the virtual try-on technology is an important application of computer vision.Therefore,the aim of this article is,through the research and improvement of virtual eyeglasses try-on technology,to promote the development of virtual try-on technology.Based on feature detection and through analysis of the process of eye location algorithm,this paper proposed an improved human eye location algorithm.The algorithm improves the positioning accuracy and speed.Based on the improved eye location algorithm,a multi-perspective eyeglasses try-on algorithm is proposed to realize the dynamic and realistic effect.The main work is as follows:This paper analyzes the principle of virtual eyeglasses try-on technology,the classification of the methods and the knowledge of eyeglasses in detail,focusing on the eye location process of cascade classifier.We introduces the image processing technology used in the final algorithm such as image segmentation,affine transformation,and image synthesis.This paper puts forward an improved eye location algorithm based on cascade classifier,which improves the detection speed and accuracy through taking advantage of detecting the human eye characteristics of cascade classifier and the actual situation of virtual eyeglasses try-on.The algorithm,using small amplitude of head motion and the human eye detection results of precursor frame,sets the prediction window to limit the detection range of the cascade classifier.The procedure for selected optional area is designed to obtain the pair of eyes correctly.The human eye re-detection based on gray level information is introduced in order to improve the positioning accuracy.Experiments and analysis results show that the improved eye location algorithm has better performance on positioning accuracy and detecting speed.This paper proposes a multi-perspective eyeglasses try-on algorithm for the continuous image sequence,which makes try-on effect more realistic.The algorithm first adopt the improved eye location algorithm for eye location,then get the pupil position and imaging pupil distance,and form a continuous imaging pupil distance sequence,and then to correct the pupil distance mutation to reduce the detection error of imaging pupil distance.Through the establishment of the head and the camera projection model to estimate head posture accurately.Finally,the eyeglasses image is fused to face image sequence to complete virtual eyeglasses try-on according to the eye position and head posture information.The experimental results show that the algorithm is dynamic multi-perspective and lifelike virtual try-on effect.