2012/07/17 by Mehrdad J. Gangeh, Gangeh, Mehrdad J., Robert P. W. Duin +5
Chemistry · Computer Science · #Computer Vision and Pattern Recognition (cs.CV) #FOS: Computer and information sciences #Image Retrieval and Classification Techniques #Machine Learning (cs.LG) #Neural Networks and Applications #Spectroscopy and Chemometric Analyses #cs.CV #cs.LG
paper · pdf · doi:10.48550/arxiv.1207.4089
28 pages
arxiv created 2012/07/17 · openalex publication_date 2012/07/17 · arxiv updated 2015/03/20 · openalex created_date 2025/10/10 · openalex updated_date 2026/07/28
Textures often show multiscale properties and hence multiscale techniques are considered useful for texture analysis. Scale-space theory as a biologically motivated approach may be used to construct multiscale textures. In this paper various ways are studied to combine features on different scales for texture classification of small image patches. We use the N-jet of derivatives up to the second order at different scales to generate distinct pattern representations (DPR) of feature subsets. Each feature subset in the DPR is given to a base classifier (BC) of a two-stage combined classifier. The decisions made by these BCs are combined in two stages over scales and derivatives. Various combining systems and their significances and differences are discussed. The learning curves are used to evaluate the performances. We found for small sample sizes combining classifiers performs significantly better than combining feature spaces (CFS). It is also shown that combining classifiers performs better than the support vector machine on CFS in multiscale texture classification.