Code covered by the BSD License  

Highlights from
Objective evaluation of binarization methods for document images

Objective evaluation of binarization methods for document images


Reza Farrahi Moghaddam


18 May 2010 (Updated )

Several measures are implemented to evaluate the output of the binarization methods.

objective_evaluation_core(u, u0_GT, u_SKL_GT)
function temp_obj_eval = objective_evaluation_core(u, u0_GT, u_SKL_GT)
% This function can be used to evaluate objectively the performance of binarization methods for document image.
% December 27th, 2012, By Reza FARRAHI MOGHADDAM and Hossein ZIAEI NAFCHI, Synchromedia Lab, ETS, Montreal, Canada
% May 17th, 2010, By Reza FARRAHI MOGHADDAM, Synchromedia Lab, ETS, Montreal, Canada
% The implemented measures are as follows [1]:
% Precision: 
% Recall: 
% Fmeasure: (used as one of the measures in "Document Image Binarization Contest" (DIBCO'09) in ICDAR'09)
% Sensitivity: (the same as Recall)
% Specificity:
% BCR: The balanced classification rate
% AUC: (The same as BCR)
% BER: The balanced error rate
% SFmeasure: F-measure based on sensitivity and specificity
% Accuracy: 
% GAccuracy: Geometric mean of sens and spec (to be used as the measure in "Quantitative evaluation of binarization algorithms of images of historical documents with bleeding noise" contest in ICFHR'10)
% pFMeasure: pseudo F-Measure
% NRM: Negative rate metric
% PSNR: Peak signal-to-noise ratio
% DRD: Distance reciprocal distortion metric [2]
% MPM: Misclassification penalty metric [3]
% [1] M. Sokolova and G. Lapalme, A systematic analysis of performance
% measures for classification tasks, Information Processing & Management,
% 45, pp. 427-437, 2009. DOI: 10.1016/j.ipm.2009.03.002
% [2] H. Lu, A. C. Kot, Y. Q. Shi, Distance-Reciprocal Distortion Measure
% for Binary Document Images, IEEE Signal Processing Letters, vol. 11, 
% no. 2, pp. 228-231, 2004.
% [3] D. P. Young, J. M. Ferryman, PETS Metrics: On-Line Performance 
% Evaluation Service, ICCCN '05 Proceedings of the 14th International 
% Conference on Computer Communications and Networks, pp. 317-324, 2005.%
% temp_obj_eval = objective_evaluation_core(u, u0_GT, u0_skl_GT);
% where
%       u is: the input binarized image to be evaluated.
%       u0_GT: is the ground-truth binarized image.
%		u0_skl_GT: is an optional input for the ground-truth skeleton of u0_GT. If not specified, the skeleton is automatically calculated using the thininng method.
%       temp_obj_eval: is the output. The measures can be reached as the fields of temp_obj_eval. For example:
%           fprintf('Precision = %9.5f\n', temp_obj_eval.Precision);

% get Precision and others
if (nargin == 2)
    temp_obj_eval = my_compute_precision_recall_fmeasure(u, u0_GT);
    temp_obj_eval = my_compute_precision_recall_fmeasure(u, u0_GT, u_SKL_GT);



Contact us