Pre-segmentation of high resolution images thanks to the morphological pyramid

Abstract

Very high spatial resolution images are now avail- able (about 1m), and processes such as classification or seg- mentation are less efficient on this kind of images. This paper proposes a method for pre-processing images, which smoothes the heterogeneous areas but perserves the borders of main objects. This process is done by 1. a multi-resolution non- linear decomposition with a morphological pyramid and 2. a recomposition step. An example is shown on a SPOT5 image, with a 2.5m spatial resolution. I. INTRODUCTION In remote sensing, the improvement of the technology leads to the arrival of a new class of optical images, having very high spatial resolution (1 m for the IKONOS satellite; 0.66 m for the QuickBird satellite and 2.5 m for the SPOT 5 satellite). Future sensors will reinforce this tendency. Consequently, the exploitation of such images offers new capabilities of applications. However, some of the methods currently used for the processing of remotely sensed images become less efficient. This phenomenon is a consequence of the important heterogeneity that appears when the spatial resolution is less or equal than one meter: at coarser resolutions, characteristic objects of an image (fields, rivers, forests, ...) are globally homogeneous, and processes based on this criterion are rather efficient (classification, segmentation, ...). This is not the case on very high resolution images. In this paper, we focus on the difficulty of segmenting such high resolution images and we propose a new method for improving the result. The solution is to insert a pre-segmentation process, based on the principle of the morphological pyramid, prior to the processing. The morphological pyramid is a multi-resolution decom- position and reconstruction using morphological filters. The following processing steps are iterated to perform the decom- position: • low-pass filtering based on morphological filters, • sampling, • calculation of details images. Then, an exact reconstruction is available. The use of morphological filters enables the extraction of details with regards to their structures within the image.