CONTENT-BASED IMAGE RETRIEVAL

Abstract

In this research, an image retrieval algorithm is studied and implemented. The salient concept of our strategy is using input color contents and texture information as the retrieval indexes. The query can be an image example or simply issued with an interactive color and texture picker. Potential applications include photo-journalism, art and fashion etc. In certain cases, users may want to retrieve some images similar to the one that they had at hand. In some other cases, users just want to browse some images that have certain specified colors and/or texture. e.g. a fashion designer may want to review those fashion patterns with blue and white as the major colors. For color similarity, the comparison of normalized color histograms between query and the images in the image base is adopted. For texture similarity, both coarseness and contrast measurements will be evaluated. To facilitate the browsing of retrieved images, a 'quick-look' version of each image is created. The size of those 'quick-look' images is around 100 pixel X 100 pixel such that more retrieved images can be displayed simultaneously while the output resolution is still acceptable. A prototype system is developed in MS-DOS environment and is written in C language. The test image base contains about 260 images with various categories. In this thesis, we will present the main algorithm for the retrieval strategy as well as the example query results. Moreover, directions for further research will be discussed.