To schematize images through texture segmentation

In Content-Based Image Retrieval (CBIR), images have to be segmented into the main semantic parts describing the objects present in the image. The texture is one of important features which identify and characterize the objects of an image. So, we propose a method based on wavelet decomposition. This technique delivers a schematic description of an image : that is to say, it provides the main regions generally associated to the objects of the images. The process divides images into (2 × 2) blocks. Then it tracks texture through levels of wavelet decomposition to form texture vectors. When vectors are calculated, a Kmeans algorithm partitions these vectors into clusters and a connected component extraction delivers the final schema. Experiments on real images are exhibited.