Research On Ear Recognition Based On Improved Sparse Representation

Biometric recognition technology refers to a technology to complete personal identification through the physiological or behavioral characteristics of the human body.As a branch of biometrics,ear recognition technology has attracted extensive attention from scholars at home and abroad in recent years because of its unique physiological location and structural characteristics.However,in a complex environment,the ear image is disturbed by various factors such as occlusion,noise,and posture transformation,which makes the ear recognition effect not ideal.Therefore,this paper studies the ear recognition method in complex environment,and proposes a fast low-rank recovery sparse representation based on H-G feature.First,preprocessing operations such as grayscale,scale normalization and image enhancement are performed on the ear image to make the processed image more suitable for subsequent work.Study the theory of sparse representation classification(SRC),introduce the research method of ear recognition based on SRC,and conduct simulation experiments to initially verify its feasibility.Secondly,considering that the traditional SRC uses the unit matrix as an error dictionary,it can not describe the problem of human ear image errors in complex environments.In-depth study of low-rank recovery matrix algorithm(LRR),and improve on it to get fast low-rank matrix recovery algorithm(FLRR).Combining FLRR and SRC,propose a ear recognition method based on fast low-rank recovery sparse representation(FLRR-SRC),which improves the recognition effect of traditional SRC in complex environments and overcomes the decline in recognition rate caused by insufficient training samples.Finally,aiming at the problem of the single feature's insufficient ability to describe the ear image in complex environments,study the feature extraction method combining Gabor feature and HOG feature(G-HOG).An ear recognition algorithm based on G-HOG features for fast low-rank restoration sparse representation is proposed.Simulation experiments show that when the ear is in a complex environment,the algorithm has a good recognition effect and is robust to interference such as occlusion,attitude deflection,and noise.