Data Compression and Decompression of Handwritten Digital-Ink Using Sparse Gaussian Process

In this paper, we address the problem of compressing and decompressing handwritten digital-ink data, comprising N points, obtained from electronic devices such as pen tablets, etc. We propose a novel method for data compression and decompression using sparse Gaussian process regression, which allows us to compress the digital-ink data into a set of $M(\ll N)$ pseudo inputs. Furthermore, we demonstrate that adjusting the hyperparameters of the kernel function during the decompression process enables the conversion of digital-ink style to a cursive form, providing enhanced flexibility and customization in the resulting digital-ink. Additionally, we show that the blurring of brush strokes can be represented by generating and superimposing characters from a Gaussian process, based on the mean and variance obtained. The performance of our proposed method is demonstrated through a series of experimental studies.