Research On The Effects Of Compression Techniques For Human Ear Recognition

Biometric recognitions have been widely applied in surveillance applications,forensics and criminal investigations.Since biometric system can provide much higher security solution than traditional personal authentication systems such as tokens or passwords,where tokens can be stolen,the long passwords or secret codes are difficult to remember and can be forgotten.Moreover,with the increase in need for more security systems in forensics and in the fields of security such as access control,immigration,and commercial applications,biometric systems have attracted much attention recently.Recently,ear print has received considerable research interests in biometric community due to its several prominent advantages.Human ears are large and visible for acquisition,stable through age and expressions,and are different for identical twins and triplets.With the increasing interest of usage ear biometric system as face and fingerprint biometric systems in many applications especially on surveillance and forensics,its need to compress the ear image data,the ear image data may be transferred through a network to a specific root or location.Due to the limitation of storage capacity and transfer data,may be over low quality wireless channels,the applications should be taken in considers the effect of image compression on its systems.However,there is no previous works present the effects of compression techniques on ear biometric system.Therefore,this dissertation firstly study and analysis the effects of known compression algorithms(JPEG,JPEG2000,and BPG)on the ear recognition system especially on two public and available ear databases.Recently,JPEG and JPEG2000 standard have played a vital role on biometric applications such as face and finger-print biometrics.Therefore,the first proposed approach for this work study the impact of JPEG and JPEG2000 on ear recognition performance which the ear image may transfer over low quality wireless channels.Firstly,local ear descriptors such as POEM,LPQ,LBP,and BSIF are used to represent and extract ear image features.Then,JPEG and JPEG2000 are exploited to compress the ear images.Then,support vector machine(SVM)is adopted for ear recognition at various compression rates.To evaluate our study,the experimental results are conducted on two public ear databases;USTB and IIT Delhi databases.The results show the superior of LPQ descriptors till compression rate 0.4 bpp.After that,to improve our previous work in feature extraction and image compression,we investigate a novel approach for compression ear recognition system based on Convolutional neural networks(CNNs)for feature extraction to represent ear images and Better Portable Graphic(BPG)for ear image compression.Our motivations on improving for representation ear images and exploiting the advantages of BPG algorithm for ear image compressions.The proposed approach adopts BPG algorithm to compress testing ear image firstly.Then,we exploit VGG-M pre-trained to extract training and testing ear deep features.Finally,SVM is used for ear classification.The experimental results show the improved of the proposed approach than previous works.Moreover,the proposed approach can achieve promising performance for compression ear biometric system.