IJMTES – SCENE LABELING SCHEME BASED HIERARCHICAL FEATURES AND SIMILARITY ANALYSIS

Journal Title : International Journal of Modern Trends in Engineering and Science

Author’s Name : T.Chandrasekar, N.Magendiran

Volume 01 Issue o5  Year 2014  

ISSN no:  2348-3121  

Page no: 128-134

Abstract—Scene labeling process assigns class label for each pixel in an image with the category of the object. Full scene labeling is referred as scene parsing process. Images are analyzed with edge and color values. Pixel information’s are used in the color property extraction. Texture and contrast are pixel based features. Classification techniques are used to assign labels to the images and detection, segmentation, multi-label recognition operations are involved in the scene parsing process. In proposed system, to perform tracking in an illumination insensitive feature space called the gradient logarithm field (GLF) feature space. Convolutional Network (ConvNet) maintains the picture information with associated labels. Filter bank module, nonlinearity and a spatial pooling module are used in the ConvNet process. Multi-Scale Convolutional Network (MSCN) is trained from raw pixels to extract dense feature vectors and encodes regions of multiple sizes centered on each pixel. Texture, shape, and contextual information are used in the scene labeling process. Multi-scale invariant features are learned from the labeled images and  super pixel is identified with the similar pixel color information in the same region. Conditional Random Field (CRF) is trained to produce labels for each candidate segment and to ensure that the labeling is globally consistent. Multiple post processing methods are used to produce the final labeling and this system automatically retrieves the best explained scene from a pool of segmentation components. The hierarchical feature selection based scene labeling scheme is improved with objective functions. The graph construction process is improved with nonlinearity rectification and image contrast normalization process. The graph cut segmentation technique is integrated with the hierarchical feature learning scheme from pattern analysis. Feature similarity analysis is performed with shortest path discovery mechanism on neighborhood graphs.

Keywords—Label, Gradient logarithm field, Convolution Network, Non-linear Rectification

Reference

[1] C. Liu, J. Yuen, and A. Torralba, “Nonparametric Scene Parsing: Label Transfer via Dense Scene Alignment,” Artificial Intelligence, (2009).
[2] D. Munoz, J. Bagnell, and M. Hebert, “Stacked Hierarchical Labeling,” Proc. 11th European Conf. Computer Vision, Jan. (2010).
[3] V. Lempitsky, A. Vedaldi, and A. Zisserman, “A Pylon Model for Semantic Segmentation,” Proc. Advances in Neural Information Processing Systems Conf., (2011).
[4] Clement Farabet, Camille Couprie, Laurent Najman, and Yann LeCun “Learning Hierarchical Features for Scene Labeling” IEEE Transactions on Pattern Analysis and Machine Intelligence, Vol. 35, No. 8, August (2013).
[5] J. Carreira and C. Sminchisescu, “CPMC: Automatic Object Segmentation Using Constrained Parametric Min-Cuts,” IEEE Trans. Pattern Analysis and Machine Intelligence, vol. 34, no. 7, pp. 1312-1328, July (2012).
[6] C. Russell, P.H.S. Torr, and P. Kohli, “Associative Hierarchical CRFs for Object Class Image Segmentation,” Proc. IEEE Int’l Conf. Computer Vision, (2013).
[7] D. Grangier, L. Bottou, and R. Collobert, “Deep Convolutional Networks for Scene Parsing,” Proc. Int’l. Machine Learning, (2012).
[8] H. Schulz and S. Behnke., “Learning Object-Class Segmentation with Convolutional Neural Networks.”Proc. 11th European Symp. Artificial Neural Networks, (2012).

Full Pdf Paper-Click Here

Scroll Up