Content-based
search engines are gaining popularity with the growth of multimedia resources on the Internet. They offer the potential for visual search, by which Internet users can specify their needs and make selections based on images, rather than text. In this project, a
Content-based image search and retrieval system that explores the
colour feature of images has been proposed. The system allows users to load a query image into the system and then searches the image
database for all images with similar colour compositions to the query image. Our image database consists of about 1000 JPEG images. Three different colour representation techniques have been used and a comparison of their retrieval effectiveness is made in terms of the recall and precision metrics. The first method is the global colour histogram method where image classification is performed according to the global colour feature. The second method used is the sub-image histogram method. In this approach, spatial colour information is provided by dividing images into equal-size segments and computing colour histogram for each segment separately. In the third technique, only the dominant features of the colour distribution are computed and stored in the database, i.e, the average colour, variance and skewness of each colour channel. This technique
results in faster retrieval time and requires smaller storage space. The results of various experiments are presented and used as a basis for evaluation of the different approaches and metrics used. The major conclusion is that, the methods working with the colour histograms produce satisfactory results though at times, as any colour indexing methods they are bound to retrieve false positives. Relatively, better results are obtained when considering spatial location of colours in images using the sub-image histograms. A comparison of the retrieval effectiveness of all three methods shows that the method that works with the moments produces more accurate results.
http://razvi.wordpress.com/