The current methodology to detecting similar fonts is based on DeepFont feature vectors. For example, let’s call query font A, and the one being scanned for similarity font B. This methodology works by finding the latent representation of both font A and B, then taking the Euclidean distance between latent presentations to find out the similarity score. Variable fonts are the latest addition to the OpenType format. A variable font is a single file that acts like multiple fonts. By changing the axis of a variable font, a new instance of the variable font is created and behaves like a normal font. Theoretically, there may be an infinite number of such instances. As there are an infinite number of combinations possible, the current methodology of finding similar fonts will fall short if the font to be matched is a variable font. We propose a methodology which uses “DeepFont” technique and enhances its result by auto-adjusting the visual property of variable font to find the instance that should be most similar to the query font A. Then, defining a greed-based approach to detect the most similar instance of font B as a variable font. There are two key aspects of this proposal: 1. Finding the similar instance of a variable font to a non-variable font. 2. Intelligently identifying the design axes of a variable font that are relevant to computing similar instances of a variable font, and 3. The use of DeepFont technique to compute similarity score between instance of variable font and non-variable font.
Efficient Methodology to Detect Similar Instance of Variable Font
Arushi Jain
