Impact of intraband misregistration on image classification

Abstract

Remote sensing data acquired from spaceborne platforms in multispectral channels with moderate to high spatial resolution has been extensively used for numerous applications. Registration between images as well as multispectral bands significantly affects the classification accuracy. Data acquired in multiple channels needs accurate intraband registration to minimise classification errors. Availability of very high spatial resolution data such as from SPOT, IRS-P6, IKONOS, and Quickbird demands very accurate intraband registration. Ability to provide accurate intraband registration requires proper knowledge of satellite attitude, Earth rotation correction, sensor geometry etc. While every effort is made to minimise the intraband misregistration at product generation level, it is difficult to remove it all together. In view of this and its significance on remote sensing image classification, an attempt was made to evaluate the impact of intraband misregistration on classification of remote sensing image with high spatial resolution data. Study carried using a prototype image and IRS-P6 LISS-IV image reveals that image data with intraband misregistration greater than 20% significantly reduce image sharpness and leads to misclassification. Though misregistration of NIR band has major impact on classification it was also seen that misregistration among all bands would lead to even greater error in classification and increased edge blurring.