Principal Components and Invariant Moments Analysis-Font Recognition Applied

摘要

An alternative Font Recognition (FR) methodology is proposed in this work; this methodology is based on the global texture analysis of document text using invariant moments (invariant to scale, rotation and translation) and Principal Components Analysis. There is not necessary of confined analysis for each letter or word. In our method, the central moment features are extracted as a global characteristic from each font (from window estimation) and over the same window, the principal components analysis is applied. An uniform text block with a unique font is suitable to provide the specific font properties necessary for the process of recognition. The proposed scheme were tested with the following fonts: Courier, Arial, Bookman Old Style, Franklin Gothic Medium, Comic Sans, Impact, Modern and Times New Roman; and their respective styles: regular, italic, bold, italic with bold. The invariant moment technique is used in this study to extract the font characteristics by window size estimation; from an entry text set a data base was build for the learning stage, and then standard statistical classifiers were applied for the identification stage. We found that the invariant moments combined with principal component analysis give excellent results in font identification task. An exhaustive study was performed with 8 types of fonts commonly used in the Spanish (Mexican) language. Each type of font can have four styles leading to 32 font combinations total, additionally, our results include tests over different font sizes: 6, 8, 10 and 12 points and different orientations: 0, 45, 90 and 135 degrees The robustness of the algorithm is also examined in terms of Gaussian noise.