An approach to image retrieval using Distance-Angle Pixel Array

Feature extraction is one of the most important issues for designing content-based image retrieval systems. In this paper, we present a method of extracting feature to retrieve images. In the first step of the proposed method, bound is introduced to mark the contour of the object. The center and the circumcircle of the object can be obtained from the bound. The principal angle of the object is computed and the object is rotated at the principal angle thereof. The included angle, made by a horizontal line and the line connecting each point within the rotated object and the center of the circumcircle, is calculated. In addition, the distance from each point within the rotated object to the center of the circumcircle is computed, and the target can be divided into several regions by distance and angle. A matrix is constructed by dividing the distance horizontally and dividing the angle vertically into several regions. The value of all pixels in the rotated object contributed to the value of the corresponding region of the matrix. The matrix is transformed into a vector of Distance-Angle Pixel Array which is the feature to retrieve similar images. Experiments results of testing 1200 images show that the recognition rate becomes higher when adding the proposed method.