Reconstruction Based Target Recognition With Application To Face And Radar Target

Target recognition, which is called pattern recognition in a broad sense, became adiscipline quickly in the1960s, with the development of computer and artificialintelligence since1940s. When we say pattern, we means the information capturedfrom specific matters and it has features distributed in the space domain and timedomain. Pattern recognition is a process to classify a pattern to the right classification.Intelligent monitoring system is an important feature of our society, we can receive avideo sequence contains lots of similar and continuous images of target by themonitoring system and it is not several discrete images any more. How to use therelevant information between similar and continuous images to improve targetrecognition rate is a remarkable problem. Inspired by super resolution images and radartarget recognition, we proposed the target recognition method using couples of testimages belonged to same class. Two categories are used in the proposed recognitionmethod, one is fusion of test images and another is fusion of residual. The experimentswere carried out on AR database, AT&T database and GT database. PCA, randommatrix and down sample images were used to reduce dimensions. Sparse representationbased classification (SRC), linear regression classification (LRC) and collaborativerepresentation based classification (CRC) were the three classifiers we used. Underdifferent conditions, we compared the result of experiments, which showed that thetarget recognition method using couples of test images can improve the recognition ratioabout5%in minimum; the maximum could reach to20%. We also applied linearregression classification (LRC) and collaborative representation based classification(CRC) to radar high resolution range profiles target recognition, inspecting theperformances of this two classification methods and comparing them with the result ofsparse representation based classification (SRC) on the MSTAR database, which is aSAR image database and the SAR images were converted into high resolution rangeprofiles (HRRP), being used to classify.