The Study of Method for Feature Extraction Based on Angle of Polarization Parameter Image

Abstract

Multispectral polarimetric imagery can provide several parameters of the man-made targets, including spectrum, intensity, state of polarization, and shapes. Among these, the state of polarization is additional useful information. It will help to improve the capabilities of the target detection and recognition. Recently, an experimental multiband polarized CCD camera has been built in our laboratory. It can obtain the polarimetric imaging information at four bands subsequently. Here, We use one kind of man-made target as an experimental sample. Many digital polarization imaging data were acquired in the field and the laboratory. Thereby the Stokes parameters I, Q and U were calculated according to the measured radiance images of three polarization directions at 0o, 60o and 120o relative orientation of the transmission optic axis of the linear polarizer. At last, the angle of polarization (AOP) image and intensity image were derived from Stokes parameters, respectively. These images were only analyzed at 443 nm waveband. The median filtering and single threshold method were adopted in angle of polarization image processing. Moreover, the vector images of the target were obtained. The result indicates that the properties of the man-made target and the background were characterized clearly by using the AOP parameter. The smooth surface of man-made target shows the coincident feature in AOP image. On the contrary, the surface of grass is unorderly. Indeed, it is the very function of the AOP image, which we want to illustrate. Hence, it shows that the AOP image can describe the different of surface orientations. Also, it may enhance the contrast between the man-made target and grass background. In a word, this new method based on the angle of polarization parameter can be used to discriminate the target from the background when we interpret the image of the polarized remote sensing.