Camouflaged Foreground Object Detection

Abstract

Automatic target detection is one of the potential research area for an adaptive solution in various inputs like: camouflaged images, IR Images, FLIR images and images with small targets. Detection of a camouflaged object is a continuous research problem in many computer vision applications where similar intensity information among foreground and background regions. An Image consists of three types of functional information like: Spatial, Spectral and temporal, the camouflage problem cannot be solved by having these three types of functional information or even together. Conventional objective and subjective methods have given a sort of solution to distinguish foreground and background but the difficulty is observed where ever the colour entity information is very close i.e: very small difference between FG and BG. This paper proposed a framework with a probabilistic approach to classify FG and BG using principle features at each pixel. Certain basic features of an image by the background representation is then modeled by significant basic features under the framework at each pixel of basic features. The characterized background approach will be used to detect the statistical information of foreground measurements. The proposed method in comparison with survey methods found to be satisfactory and can be useful in computer vision applications, texture classification, military, and civilian, etc.