Spiral Representation Of Spatial Feature for Spatial- Color Based Image Retrieval Wang Hui Hui 1 , Narayanan Kulathuramaiyer 1 , Wang Yin Chai 1 , Abdelhamid Abdessalam 2 1. Faculty of Information Technology, Universiti Malaysia Sarawak 2. Department of Computer Science, University of Qatar, Qatar Email : hhwang@fit.unimas.my, nara@fit.unimas.my, ycwang@fit.unimas.my, ahamid@qu.edu.qa ABSTRACT In this paper, we present a new approach for retrieving similar images based on color-spatial feature. We propose the Spiral Representation Spatial Feature for images, which is rotation and scaling invariant. The Spiral Representation Spatial Feature is based on bit signature approach. The feature is extracted based on the Spiral Extraction Method, where the sub area is extracted in a circular manner, starting from the inner square and read in counter clockwise direction. The starting point of every square is the left upper sub area. The derivation of the rotated spatial features is proposed to be used for retrieval of rotated images. As an additional to common matching of whole images, partial image matching has been introduced to compliment the whole image matching, where the same images with different sizes and scaled, are retrieved as similar images, and also to locate the similar image appeared as sub image in another whole image. I. INTRODUCTION Information over flow has become widespread with the explosive advancement in imaging technologies. Image retrieval is attracting interests among researchers in the field of image processing, multimedia, digital libraries, remote sensing, astronomy, database applications and others related area. The effectiveness and efficiency of the image retrieval system is able to operate on the collection of images to retrieve the relevant images based on the query image that conform to human perception. By identifying and mapping an image based on color, texture and shape alone is not sufficient to produce satisfactory and accurate results. A major class of users’ requests require retrieving those images in the database that are spatially similar to the query image[3]. Hence, the spatial information recently has been addressed in the field of image retrieval. However, spatial feature can be used to integrate with other visual features to better capture the contents of images. This will greatly improve the accuracy in image retrieval and at the same time producing results that is conform to human perception. To facilitate more effective and efficiency image retrieval and also to meet and fulfill the major class of users’ requests, some integrated color-spatial approach has been explored in our literature [5, 7, 8, 9, 10, 11]. These approaches are having some weaknesses especially they are either only rotation invariant in certain degree and certain class of image or not rotation and not scaling invariant (not able to perform partial image retrieval) at all. By introducing our spiral representation spatial feature that are rotation and scaling invariant, we manage to overcome those rotation and scaling invariant problems and at the same time we also manage to archive accurate and effective results in retrieving similar or related images. Next discussion focuses on the results from the reviews of some related works on spatial-color information image retrieval and also the proposed spatial-color feature extraction and representation based on spiral feature. II. RELATED WORKS This paper highlights several spatial-color based image retrieval works. Bit signature approach, histogram approach as well as the clustered approach has also been reviewed. Chua, Tan, Ooi[9] carried out their signature based approach by extracting the dominant color of the image as representative color. The image is then partitioned into m x n cells of equal size to get the color and spatial distribution. The cells that have the representative colors are mapped into bit position of color signature. So, the image representation consists of a list of all bit signature of the representative color. The weaknesses of this approach is that they cannot effectively handle for those images with no dominant colors and images those dominant colors are scattered. It is also totally not rotation and scaling invariant. So, it cannot retrieve the rotated and scaled image of the query image from database effectively. Jo and Um[8] produce better result than Chua, Tan, Ooi[9]’s approach by proposing two axial signature- based method. The Dominant Color Composition (DCC) that is created with regionally representative