Blind Separation Of Scanned Documents Mixture Based On Nonlinear Preprocessing and Competition Between The High-Frequency Components

Arbabi, Somaye; Ghaderi, Reza; Ebrahimzadeh, Ataollah

Blind Separation Of Scanned Documents Mixture Based On Nonlinear Preprocessing and Competition Between The High-Frequency Components

Document Type : Persian Original Article

Authors

Somaye Arbabi

Reza Ghaderi

Ataollah Ebrahimzadeh

Department Of Electrical And Computer Engineering, Babol Noshiravani University Of Technology, Iran.

Abstract

When an image is taken from a paper-based documents, such as scanned paper, the picture on the back page of the front page of the paper leaves effect behind. Pictures of the back and the front page of the paper, make a nonlinear combination and thus a practical problem of blind source separation (BSS) provides. Here a version of the highly nonlinear mixtures using onion skin paper, is used. Separation was performed by applying an enhanced InfoMax algorithm based on a novel nonlinear preprocessing method, and competition between high frequency components of the wavelet decomposition of the images. Separation results with proposed algorithm were evaluated by objective and subjective criteria. that indicate the precision, speed and quality of the estimated image with proposed algorithm, outperform the other nonlinear BSS methods. In order to accurately evaluate the separating algorithms and comparison the quality of estimated images with original sources, in addition to the usual criteria, the similarity Line Profile and measure of Euclidean distance between standard deviation of images introduced. Nonlinear separation results with the proposed algorithm was tested by the quantitive and qualitative measures, and a significant improvement was observed.

Keywords

Blind Source Separation

Nonlinear mixture of images

Independent Component Analysis

Enhanced InfoMax

wavelet transform