From Images to Schemas

In content based image retrieval (CBIR), images are segmented to synthesize image information. Among several characteristics like color or edges, texture is useful for segmenting. This paper proposes an intensive multiresolution approach to texture segmentation based on a wavelet transform. The technique delivers schematic descriptions of images. That is to say, it provides the main regions of interest (ROIs) according to image information. Firstly, the process divides images into 2 times 2 blocks. Then, it tracks texture through the multiresolution offered by the wavelet transform to form featuring vectors. Next, a K-means algorithm partitions the texture vector space into clusters. Finally, a connected component extraction delivers the image schema.