Robustness of shape descriptors to incomplete contour representations

Abstract

Abstract-With inspiration from psychophysical researches of the human visual system, we propose a novel aspect and a method for performance evaluation of contour-based shape recognition algorithms regarding their robustness to incompleteness of contours. We use complete contour representations of objects as a reference (training) set. Incomplete contour representations of the same objects are used as a test set. The performance of an algorithm is reported using the recognition rate as a function of the percentage of contour retained. We call this evaluation procedure the ICR test. We consider three types of contour incompleteness, viz. segment-wise contour deletion, occlusion, and random pixel depletion. As an illustration, the robustness of two shape recognition algorithms to contour incompleteness is evaluated. These algorithms use a shape context and a distance multiset as local shape descriptors. Qualitatively, both algorithms mimic human visual perception in the sense that recognition performance monotonously increases with the degree of completeness and that they perform best in the case of random depletion and worst in the case of occluded contours. The distance multiset method performs better than the shape context method in this test framework.

Confidence interval approach to feature re-weighting

“…11 Confidence intervals with no overlap Fig. 10 Possible cases for confidence intervals in the negative range ub 1 þub 1 lb 1 þ ub 1 x < ub 2 þub 2 lb 1 by substituting (1) ub 1 þub 1 lb 1 þ ub 1 x < ub 1 þy þ ub 1 lb 1 þ lb 1 y by substituting (2) ub 1 x < y þ lb 1 y 0 < y þ lb 1 y À ub 1 x 0 < y 1 þ lb 1 ð ÞÀub 1 x is true considering (3), (4), (5), (6).…”

“…10b and e) Figure 12 illustrates the case where there are two confidence intervals that have partial overlap. Let (2) ub 1 x < y þ lb 1 y 0 < y þ lb 1 y À ub 1 x 0 < yð1 þ lb 1 Þ À ub 1 x is true considering (3), (4), (5), (6).…”

“…Similar to the intra-feature reweighting approach of MARS, these feature re-weighting approaches cannot support features that cannot be represented in fixed-length vectors. The similarity between images using features such as [6,10,17,18] which have representations of varying length are calculated by performing a best-fit pair-wise matching approach. Best-fit pair-wise matching algorithms attempt to find the closest matching of pairs between the components of two feature representations, then compute a distance metric using a function that measures the quality of matches.…”

“…This assumption is acceptable when using features such as colour histograms since the number of histogram bins for each image is the same. However, this viewpoint is invalidated by features such as the dominant color descriptor [17], curvature scale space [18], and incomplete contour representations [6], which have representations that may differ in size between images. High-level features that are currently being researched, such as 9D-SPA [10], which describe the spatial relations between objects in an image also invalidate the assumption that features can be represented in fixed-length vectors since the number of objects may differ between images.…”

Recognition of Simple 3D Geometrical Objects under Partial Occlusion

“…In our approach we describe the global structure by a set of points of interest (POI), which we later interpolate with B-splines. Here we use a common way to define a point of interest as a local maximum of the curvature function [5,2]. Let x and y be functions defining a discrete contour in a parametric representation.…”