Corresponding author: salukak@pdn.ac.lk 1 TOWARDS THE ENHANCED FEATURES FOR CONTENT BASED IMAGE RETRIEVAL 1 S. R. Kodituwakku 2 S. Selvarajah 3 U. A. J. Pinidiyaarachchi 1, 3 Department of Statistics & computer science University of Peradeniya, Sri Lanka 2 Postgraduate Institute of Science, University of Peradeniya, Sri Lanka ABSTRACT Several low-level features such as colour, texture and shape are used in Content Based Image Retrieval (CBIR) systems. This paper presents two enhanced features for CBIR. The first method enhances color moments by taking edge information into account. Colour distribution of image pixels are analyzed for three types of edges: horizontal edge, vertical edge and one non-directional edge. A distance measure is computed based on the colour distribution of each edge type. Then the resultant distance measures are combined to obtain a similarity measurement for retrieval. The second feature is based on the combination of color coherent vector (CCV) and color moment properties of images. In this method, the pixels within the three color buckets (R, G, and B) are classified as either coherent or incoherent based on the area covered by it. Then the first, second and third order moments are separately computed for coherent and incoherent pixels. The methods have successfully been tested with an image database that consists of 1000 images of different categories. Experiment results show that the proposed features are more effective for CBIR than the individual features. Keywords: Image Retrieval, Colour moments, Directional edges, Non-directional edges, Colour Coherent Vector. 1. Introduction Content Based Image Retrieval (CBIR) has been an active research area in the last two decades. Traditional systems represent image contents only by text annotations and this approach has several limitations. CBIR combines digital image processing techniques with database techniques to retrieve images efficiently. In CBIR systems, images are automatically indexed by summarizing their visual features. A feature is a characteristic that can capture a certain visual property of an image either globally for the entire image or locally for regions or objects. Color, texture and shape are commonly used features in CBIR systems. Although some new methods such as the relevance feedback have been proposed to improve the performance of CBIR systems, low-level features still play an important role 1-5 . A key function in any CBIR system is the feature extraction. Mapping the image pixels into the feature space is known as feature extraction. Extracted features are used to represent images for searching, indexing and browsing images in an image database. Use of feature space is more efficient in terms of storage and computation.