Content-based image retrieval: a new promising technique in powder technology

The aim of the present study was to introduce a new technique for analyzing powders by examining the content information of images of pharmaceutical powder systems. Texture features of images of microcrystalline cellulose were compared by using a content-based image retrieval system (CBIR), QBIC (Query-by-Image-Content). The rank order and image similarities were compared to particle sizes and appearances of different mixtures. The image order of the similarity values was in close agreement with the appearance and particle size of the mixtures. When the image of pure Avicel PH 101 was used as a query image, the most similar images were always from images of mixtures with a large number of particles with smaller particle mean sizes. When images of pure Avicel PH 200 were used as a query image, the closest matches of image similarity were from images of mixtures with a larger amount of larger particles. The results show that the CBIR system extracts applicable content information on images of powders, but the texture features used were not totally adequate for analysis of the powders used. In general, content-based image retrieval seems to be a promising approach to efficiently use the vast image information that is available from pharmaceutical powders. Nevertheless, to achieve an efficient CBIR tool for powder technology requires development of substantial algorithms for feature extraction.