A modified RLSA technique for automatically reading trademarks

Abstract

This paper proposes a technique for automatically reading trademarks that have interflow of mixed-mode contents of text and nontext. The design is based on die known knowledge of the structure of the trademarks. First of all, block segmentation using modified Run Length Smoothing Algorithm (RLSA) decomposes the trademark image into single-mode blocks. Then, group classification classifies the blocks into various groups followed by classification of text and nontext block. Then, extraction process of the text block proceeds using the average length of the runs of the white pixels. The experimental results show that the proposed technique is capable to perform text extraction for the trademarks that satisfy the requirements.