Appearance-based methods Outline of object recognition

1 appearance-based methods

1.1 edge matching
1.2 divide-and-conquer search
1.3 greyscale matching
1.4 gradient matching
1.5 histograms of receptive field responses
1.6 large modelbases

appearance-based methods

use example images (called templates or exemplars) of objects perform recognition


objects different under varying conditions:

changes in lighting or color
changes in viewing direction
changes in size / shape




a single exemplar unlikely succeed reliably. however, impossible represent appearances of object.

edge matching

uses edge detection techniques, such canny edge detection, find edges.
changes in lighting , color don’t have effect on image edges
strategy:




measurements:

good – count number of overlapping edges. not robust changes in shape
better – count number of template edge pixels distance of edge in search image
best – determine probability distribution of distance nearest edge in search image (if template @ correct position). estimate likelihood of each template position generating image



divide-and-conquer search

strategy:

consider positions set (a cell in space of positions)
determine lower bound on score @ best position in cell
if bound large, prune cell
if bound not large, divide cell subcells , try each subcell recursively
process stops when cell “small enough”




unlike multi-resolution search, technique guaranteed find matches meet criterion (assuming lower bound accurate)


finding bound:

to find lower bound on best score, @ score template position represented center of cell
subtract maximum change “center” position other position in cell (occurs @ cell corners)


complexities arise determining bounds on distance

greyscale matching

edges (mostly) robust illumination changes, throw away lot of information
must compute pixel distance function of both pixel position , pixel intensity
can applied color also

gradient matching

another way robust illumination changes without throwing away information compare image gradients
matching performed matching greyscale images
simple alternative: use (normalized) correlation

histograms of receptive field responses

avoids explicit point correspondences
relations between different image points implicitly coded in receptive field responses
swain , ballard (1991), schiele , crowley (2000), linde , lindeberg (2004, 2012)

large modelbases

one approach efficiently searching database specific image use eigenvectors of templates (called eigenfaces)
modelbases collection of geometric models of objects should recognised




^ m. j. swain , d. h. ballard colour indexing , international journal of computer vision, 7:1, 11-32, 1991.
^ b. schiele , j. l. crowley recognition without correspondence using multidimensional receptive field histograms , international journal of computer vision, 36:1, 31-50, 2000
^ o. linde , t. lindeberg object recognition using composed receptive field histograms of higher dimensionality , proc. international conference on pattern recognition (icpr 04), cambridge, u.k. ii:1-6, 2004.
^ o. linde , t. lindeberg composed complex-cue histograms: investigation of information content in receptive field based image descriptors object recognition , computer vision , image understanding, 116:4, 538-560, 2012.