Segmentation Algorithms for Detection of Targets in IR Imagery (Algorithmes de Segmentation pour la Detection de Cibles sur Images IR)

摘要

This report presents a class of segmentation algorithms (segmenters) for detection of targets in IR imagery based on the single assumption that the targets possess a larger thermal signature than the background. This class of algorithms emerged as a result of efforts to improve an early segmenter devised to extract targets from IR BOFORS imagery. This segmenter proceeds according to a Single Intensity Threshold whence the name SIT Generator to designate it. The extraction record of the SIT Generator is generally excellent whenever the background, on a large-scale basis, is relatively uniform. When this condition is not met, one can use a thresholding intensity function in lieu of a fixed threshold. The SIT Generator and its variants try to cope with the background simply by partitioning the image. A more promising avenue consists in levelling the background so as to curb its ascendancy over the image. This is in essence what the Background Elimination Technique (BET) expounded herein does. Since BET can be applied either to the set of lines or columns of an image, it generates 2 images referred to as the Horizontal Fine Structure image and the Vertical Fine Structure image respectively. These images offer many possibilities for detection of targets and several of them are explicitly described in the report. (Author)