Home           Contact us           FAQs           
     Journal Home     |     Aim & Scope    |    Author(s) Information      |     Editorial Board     |     MSP Download Statistics
2012 (Vol. 4, Issue: 24)
Article Information:

Engineering and Image Classification Framework Using Multi Instance Learning with KCCA Algorithm

P. Bhuvaneswari and S. Chitrakala
Corresponding Author:  P. Bhuvaneswari 

Key words:  Clustering, correlation analysis, multiple instance learning, normalized, , ,
Vol. 4 , (24): 5429-5433
Submitted Accepted Published
March 18, 2012 April 23, 2012 December 15, 2012

Image classification is a challenging task with many applications in computer vision. Images are annotated with multiple keywords that may or may not correlated. Therefore, image classification may be naturally modelled as Multiple Instance Learning problem. The main challenge of this problem is that usually classes are overlapped and correlated. In single label classification the correlation among instance is not taken into account. In an image the instance may belongs to several classes. The correlations among different tags can significantly help predicting precise labels for improving the performance of multi label image classification. This study proposes a method Kernel Canonical Correlation Analysis (KCCA) and Multi Instance Learning for multi label image classification, for improving the performance of classification accuracy. The proposed framework comprises an input image which can be partitioned into image patches and features can be extracted. It breaks the original training set into several disjoint clusters of data. It then trains a multilabel classifier from the data of each cluster. The K means clustering is used to perform automatic instance cluster. Kernel canonical Correlation analysis can be made between disjoint clusters to know exact correspondence between image patches. Multi Instance Learning is one potential solution to address the issue of huge inter-concept visual similarity and improve the classification accuracy. The proposed approach reduces the training time of standard multi-label classification algorithms, particularly in the case of large number of labels.
Abstract PDF HTML
  Cite this Reference:
P. Bhuvaneswari and S. Chitrakala, 2012. Engineering and Image Classification Framework Using Multi Instance Learning with KCCA Algorithm.  Research Journal of Applied Sciences, Engineering and Technology, 4(24): 5429-5433.
    Advertise with us
ISSN (Online):  2040-7467
ISSN (Print):   2040-7459
Submit Manuscript
   Current Information
   Sales & Services
Home  |  Contact us  |  About us  |  Privacy Policy
Copyright © 2015. MAXWELL Scientific Publication Corp., All rights reserved