Estimating the compressibility of raster data

The raster data model is widely used in Geographic Information Systems and image processing. The continuous growth of raster data volume poses significant challenges for storage and management. Compact representations of rasters have emerged as a critical solution to address this issue, leveraging data locality to achieve efficient compression. In this context, the research community has proposed compressibility measures aiming to estimate the compressibility of data. Some measures, initially proposed for sequences, have been extended to two- and three-dimensional matrices. This work conducts an experimental analysis of measures applied to raster data compressibility estimation. The first approach applies a linearization function on raster data with matrix representation and then uses existing one-dimensional compressibility measures. The evaluation of the approach compares 1D compressibility measures with 2D measures, data compressors, Compact Data Structures (CDSs), and spatial locality estimation techniques. The results show that spatial locality, alphabet size, and noise directly influence raster compressibility, having more impact over measures like $z$, $v$, and $g$, compressors (bzip, gzip) and a CDS called $k^2$-raster. The second approach introduces ${\delta }_{\Delta }$, a 2D compressibility measure sensitive to differences within the alphabet values. Its purpose is to refine the estimation of raster compressibility. Results indicate that ${\delta }_{\Delta }$ is affected by the actual values and their frequencies, aligning with the outcomes of some specific compressors. This alignment underscores the suitability of ${\delta }_{\Delta }$ for compressibility estimation tasks closely related to those performed by such compressors.