Image retrieval using contour feature with rough set method

Abstract

Content based image retrieval (CBIR) is well-known in the field of image retrieval. It uses contents of an image from image processing and analysis to retrieve images that users were looking for from an image search. Shape based image retrieval was focused in this paper. A data mining was considered to find knowledge in the image database. Fourier descriptor is a most technique to extract contour feature of images. It was used to analyze the testing and training images in the preprocessing step. Fourier coefficients were quantized into multiple attributes and rough set theory was used to generate a rule-based system. Rough set theory is used as a data mining technique. It was compared to similarity measurement. We use 15, 984 testing image data with 71, 928 training image data in this experiment. A total usage time of rough set method is 13, 286 seconds. A total usage time of similarity measurement is 19, 365 seconds. A total usage memory of rough set method and similarity measurement are 2.8 Mbytes and 8.6 Mbytes respectively. An average precision, an average recall and an average accuracy of rough set method are 0.1297, 0.261 and 0.9971. An average precision, an average recall and an average accuracy of similarity measurement are 0.1619, 0.9651 and 0.9852. The rough set method is advantage to the usage time and the usage memory. © 2010 IEEE.