Texture Analysis for Content Based Image Retrieval

Content based image retrieval, texture analysis, texture image description, Daubechies' wavelet transform, k-means algorithm, size gradient

Content based image retrieval systems use for description of image queries and images in a database their own color, texture and shape properties. This paper focuses on the description of texture properties of images. A technique for texture image description is proposed, which consists of two steps: an analysis of particular texture samples obtained by an image uniform sampling followed by the k-means algorithm, which is applied for partitioning feature vectors from this analysis into image regions. The use of this approach is also presented, mainly in respect of natural textures with a size gradient. This technique was used in an experimental texture retrieval system too. An example of a query and a retrieval result are included as well.