An application of a gradient relaxation method to noisy infrared images.

Abstract

Image segmentation is an essential preliminary step in automatic pictorial pattern recognition and scene analysis problems. The objective of segmentation techniques is to partition an image into regions or components. The purpose of this thesis is to analyze a segmentation technique called gradient relaxation. The gradient relaxation method is a viable method in segmenting objects within an image. The gradient relaxation technique is applicable to images having unimodal distributions. This method is applied to noisy infrared images in an attempt to detect and classify the target. The method allows for an easy selection of a threshold value which may be required for other types of image processing on the image. The main issue is to examine the effectiveness of this technique applied to noisy infrared images from uncooled focal plane array sensor having unimodal distributions. The technique was able to extract the target in the image, producing a homogeneous and uniform region for most of the cases studied. A target which was fragmented into several parts because of the noise is not detectable. The technique could be implemented in hardware and applied to the inputs of a classification system for detectable objects in noisy infrared images.