Adaptive Minimum Classification Error based KISS Metric Learning for Person Re-identification
Jasher Nisa A J1, Sumithra M D2
1Jasher Nisa A J, M.Tech Scholar, Department of Computer Science and Engineering, LBS Institute of Technology for Women, Thiruvananthapuram (Kerala), India.
2Sumithra M D, Assistant Professor, Department of Computer Science and Engineering, LBS institute of Technology for Women, Thiruvananthapuram (Kerala), India.
Manuscript received on 13 August 2016 | Revised Manuscript received on 20 August 2016 | Manuscript Published on 30 August 2016 | PP: 151-155 | Volume-5 Issue-6, August 2016 | Retrieval Number: F4707085616/16©BEIESP
Open Access | Editorial and Publishing Policies | Cite | Mendeley | Indexing and Abstracting
© The Authors. Blue Eyes Intelligence Engineering and Sciences Publication (BEIESP). This is an open access article under the CC-BY-NC-ND license (http://creativecommons.org/licenses/by-nc-nd/4.0/)
Abstract: Person re-identification becoming an interesting research area in the field of video surveillance and is taken as the area of intense research in the past few years. It is the task of identifying a person from a camera image, who is already been tracked by another camera image at different time at different location. Manual re-identification in large camera network is costly and mostly of inaccurate due to large number of camera that he had to simultaneously operate. In a crowded and unclear environment, when cameras are at a lengthy distance, face recognition is not possible due to insufficient image quality. So, visual features based on appearence of people, using their clothing, objects carried etc. can be exploited more reliably for re-identification. A person’s appearence can change between different camera views, if there is large changes in view angle, lighting, background and occlusion, so visual feature extraction is not possible accurately. For solving a person re-identification problem, have to focus on “developing feature representations which are discriminative for identity,but invarient to view angle and lighting”. Recently, Minimum Classification Error (MCE) based KISS metric learning is considered as one of the top level algorithm for person re-identification. It uses VIPeR feature set as input, which contains the extracted features. MCE-KISS is more reliable with increasing the number of training samples. It uses the smoothing technique and MCE criteria to improve the accuracy of estimate of eigen values of covarience metrics. The smoothing technique can compensate for the decrease in performance which arose from the estimate errors of small eigenvalues. Here, the value of average number of small eigen values of the covarience metrics is set as a constant. So it does not work well for a large number of samples. In such situation, introduce a new method to find the value of average of such small eigen values by maximizing the likelihood function. The new scheme is termed as Adaptive MCE-KISS and conduct validation experiments on VIPeR feature dataset.
Keywords: Reidentification, Matric Learning, Covarience Matrics, Likelihood Method.
Scope of the Article: Deep Learning