Core Concepts
APRIL, a powerful intermediate filtering technique, uses raster interval approximations to efficiently identify pairs of intersecting polygons, reducing the number of expensive geometric intersection tests required.
Abstract
The paper introduces APRIL (Approximating Polygons as Raster Interval Lists), an enhanced intermediate filtering method for spatial intersection joins between polygons. APRIL improves upon previous raster-based techniques in several ways:
It simplifies the polygon approximation by using only two sorted interval lists - the A-list that captures all cells overlapping the polygon, and the F-list that captures only the fully covered cells. This avoids the complex cell type encoding used in prior work.
The APRIL intermediate filter applies a sequence of simple interval joins (AA-join, AF-join, FA-join) to efficiently identify true negatives, true hits, and indecisive pairs, without the need for expensive cell-level comparisons.
APRIL applies a lightweight compression technique to greatly reduce the space required for storing the interval lists, making them even smaller than object MBRs in some cases.
APRIL supports customization options, such as space partitioning and using different rasterization granularities for different polygons, to further tune its performance.
The paper also presents a novel, efficient one-step algorithm to directly compute the APRIL approximation of a polygon, without the need for full rasterization.
Experiments on real data show that APRIL outperforms the state-of-the-art intermediate filter, occupying 2x-8x less space, being 3.5x-8.5x faster, and reducing the end-to-end spatial join cost by up to 71%.
Stats
"The number of polygons in the datasets ranges from 3.1K to 7.1M, with an average of 25.4 to 2285.0 vertices per polygon."
"The average object MBR area ranges from 1.19E-04 to 3.95E-01."
Quotes
"APRIL approximations are simpler, occupy much less space, and achieve similar pruning effectiveness at a much higher speed."
"By applying a lightweight compression technique, APRIL approximations may occupy even less space than object MBRs."