Fast shadow removal algorithm for river garbage pollution monitoring system

摘要

In this paper, a fast shadow removal algorithm developed and implemented for river garbage pollution monitoring is reported. The algorithm is a modification of Retinex PDE shadow removal. In usual serial approach way, although giving satisfaction results in removing shadow in river image, Retinex PDE requires much computation tasks. In order to make it run faster, a parallel processing approach was applied. The parallel processing based shadow removal system had been developed using NVIDIA CUDA platform. The shadow removed images will then be processed using image segmentation. The segmentation process will separate between river and non-river (garbage) part in image. To analyze the accuracy of river and garbage separation, the percentages of garbage part in the segmented shadow removed image are compared with the percentage of garbage part obtained from ground truth image. The speed of computation between serial approach and parallel approach of shadow removal are also be compared and analyzed.