Quality control of image edge maps: a measure of comparison between automatic image edge labeling methods

GIMENEZ, J. A.; ELENA RUFEIL FIORI; ANA GEORGINA FLESIA

Córdoba

Congreso; IV CLAM 2012; 2012

UMALCA

This paper presents a methodology to assess the quality of image edge maps when Reduced Reference (RR) is available. The aim of this work is to study a quality curve based on a similarity cost between a smoothed version of the computed edge map and the reference edge map. We investigate the effectiveness of the methodology for identifying problems such as isolated edges, thick edges and noise insertion, considering three different labeling systems; Canny edge detector, Haar wavelet thresholding detector and a Hidden Markov Model scheme applied to selected undecimated Haar wavelet coefficients. We show the performance of detectors in an integrated plot of quality curves, for several experiments of real images, as well as noisy simulated ones. The analysis of such curves allows to select the best map related to several criteria, besides exact point to point matching.