Enhancement of Color Image Retrieval Capabilities: Fusion of Color with Texture Optimization

Abstract

Challenge in content based color image retrieval system lies in assigning synthetic descriptor to the image features which are economic in terms of memory required and time taken to compare query image features with features of image from database. Texture is a very important features of an image. Lots of mathematical model have been presented earlier, which describes the textural feature of the images, but most of them are textural features of gray scale images. If the same method tried to extend for color images the time and space parameter surpass the practical boundaries. The solution to this is to convert the color images to gray scales and then extract the textures, but the importance of color features in similarity measurement by a human observer cannot beover looked. So, either a new mathematical model is required to represent the color texture features or during comparison, combine the texture property and color property extracted separately. The work presented here extracts the color feature by quantizing it in color and pixel space, then finding the color dominance locally and globally. The texture feature are extracted using co-occurrence matrix method first, with a liberal value threshold the image names with similar texture are retrieve from the texture database and then as a second level of filtering, color feature is combined with texture feature and similar image are displayed. Authors have MATLAB for the logic implementation and postgre SQL 8.1 for windows, as the database to store image feature.