Bounded-Error Piecewise Linear Approximation for Sensor Data Compression

• Main Authors: Kai-Yuan Hsueh, 薛凱元
• Other Authors: Jeng-Wei Lin
• Format: Others
• Language: zh-TW
• Published: 2015
• Online Access: http://ndltd.ncl.edu.tw/handle/56354x

碩士 === 東海大學 === 資訊管理學系 === 104 === Many data compression algorithms for sensor data have been proposed in the literature. They usually attempt to find a good approximation to the hidden data model of sensor data. Lossy data compression algorithms keep only the approximation but drop the residual (difference between the approximation and sensor data), while lossless ones keep both the approximation and residual. However, there is no bound to the residual. Therefore, operations on the approximation are very limited. In this paper, we propose BEPLA, a bounded-error data compression algorithm for sensor data based on piecewise linear approximation. A parameter, ɛ, is introduced to limit the maximum error between the approximation and sensor data at any data point. We use the Minimum Description Length (MDL) framework to find the optimal ɛ so that the total bandwidth is minimal to transmit and store both the approximation and residual. The result of the experiments on Donoho-Johnstone benchmark dataset shows BEPLA is superior in both compression rate and error suppression compared to compression algorithms based on Adaptive Piecewise Constant Approximation, Discrete Fourier Transform, and Piecewise Linear Approximation.