Medial Representation Based Font Recognition Method

In this article, a method of font recognition based on the medial representation, integrated into the font recognition system based on a digital image of text is described. This system searches for similar fonts, ordered by similarity, to the font shown in the user-entered text image. The system is based on solving two machine learning problems: text recognition on an image and font recognition on a text image. To solve the first problem, we use the concept of a mathematical model of a grapheme based on a continuous medial representation of a symbol. The solution to the font recognition problem is based on the concept of the morphological width of the figure, which is also closely related to the medial representation. We propose a method for using the morphological width function to find the most similar fonts from a known database. The experiments show high accuracy of searching for the most similar fonts. For a database consisting of 2543 fonts, the accuracy is 0.991 according to the metric top@5 for correctly recognized text in the font size of 100 pixels in the image.

Font Recognition of English Letters Based on Distance Profile Features

This paper presents a system for recognizing English fonts from character images. The distance profile is the feature of choice used in this paper. The system extracts a vector of 106 features and feeds it into a support vector machine (SVM) classifier with a radial basis function (RBF) kernel. The experiment is divided into three phases. In the first phase, the system trains the SVM with different Gamma and C parameters. In the second phase, the validation phase, we validate and select the pair of Gamma and C values that yield the best recognition rates. In the final phase, the testing phase, the images are tested and the recognition rate is reported. Experimental results based on 27,620 characters glyph images from three English fonts show a 94.82% overall recognition rate.