Sub Pixel Classification of Remote Sensing Data using CIE Chromaticity Values

The colors in real world have sharp boundaries everyone knows exactly where a color starts and where it ends. When taken image of such an area the image is expressed in pixels, each pixel representing one value often one can say grey value in each band. These pixels don't express the boundaries exactly as sharp as they are in reality; in this paper, observed a transition from one color to some color other than the second color. This phenomenon is also discussed while defining mixed pixels, the pixels at boundary contain both the colors in a proportion so that the pixel appears the color different from either of two. The main aim of this paper is to extract the information from mixed pixels and subpixel swapping with the neighboring pixels to get the hidden information from the sub pixels. Sub-pixel level classification is essential for the successful description of many land cover patterns.

The perception-oriented measures such as the CIE color difference metrics and the spectral measures such as the Euclidean distance between high-dimensional vectors that measures in the first category usually define particular viewing conditions (illuminant, standard observer) whereas spectral measures on the other hand, allow comparing full reflectance spectra in an unconstrained fashion. In this paper, both perceptual and purely physical properties of the scene are considered to detect salient objects.