patch-based image classification using image epitomes
DESCRIPTION
David Andrzejewski Computer Sciences 766 Fall 2005. Patch-Based Image Classification Using Image Epitomes. Given Positive and negative example images for a certain classification (contains face, is outdoors, etc) Do - PowerPoint PPT PresentationTRANSCRIPT
Patch-Based Image Classification Using Image Epitomes
David AndrzejewskiComputer Sciences 766
Fall 2005
Problem Statement
Given Positive and negative example images for a certain classification (contains face, is outdoors, etc)
DoDevelop classifier capable of classifying new images as positive or negative
Image Epitomes
Consists of patches and mappingsPatches and mappings are learned with EMApplications in vision (de-noising,segmentation,others)
Input image A set of image patches Image epitome
www.research.microsoft.com/~jojic/epitome.htm
Image Reconstruction
Original image can then be reconstructed by mosaicing
epitome patches
www.research.microsoft.com/~jojic/epitome.htm
Recognition / Detection / Classification
Epitome of 295 face images
The smiling point
Images with the highest total posterior at the “smiling point”
Images with the lowest total posterior at the “smiling point”
www.research.microsoft.com/~jojic/epitome.htm
Approach● Construct collage of positive and negative examples● Learn the image epitome of the training collage● Find epitome patches that are preferentially mapped into the positive example images in the collage● Calculate P(patch(i)|pos/neg) for these patches (also use psuedo-counts)● Use these patches to classify new images by calculating odds ratio
Preliminary Results Training Collage
Negative Test Images
Epitome
Positive Test Images
Problems with Approach
● Difficult to incorporate new examples ● Would need to add to collage and re-learn
epitome (is there a better way?)● “Bag of words” → Spatial information discarded● Not model-based
● Pose/Illumination/Scale-variant ● Only way to handle variation is to include
training examples for various conditions
Potential Modifications● Cluster training images
● Ex: Training images w/ low vs high illumination● Discriminative patches may map exclusively to one
subset of positive images → take this into account
● Change “winner take all” for P calculations ● Consider relative probabilities of 'near matches'● Account for multiple mappings somehow
References
1. V. Cheung, B. J. Frey, and N. Jojic, Video epitomes, Proc. IEEE Conf. Computer Vision and Pattern Recognition, 2005
2. N. Jojic, B. J. Frey, and A. Kannan, Epitomic analysis of appearance and shape, Proc. 9th Int. Conf. Computer Vision, 2003
3. R. Fergus, P. Perona, A. Zisserman, Object Class Recognition by Unsupervised Scale-Invariant Learning, Proc. of the IEEE Conf on Computer Vision and Pattern Recognition, 2003
Testing images from Google Images and Flickr